Quick Guide to Geomorph v.2.0 Emma Sherratt (emma.sherratt@gmail.com)
[document version 1: 14 April 2014]
Table of Contents Table of Contents ...................................................................................................................................... 2 Introduction .............................................................................................................................................. 4 What is geomorph? ............................................................................................................................................. 4 How to use this manual....................................................................................................................................... 4 Workflows for common analyses ....................................................................................................................... 6 Example datasets in geomorph ........................................................................................................................... 7 Permutation tests ................................................................................................................................................. 8 Section 1: Data Input: Importing landmark data ........................................................................................ 9 1.1 TPS files (readland.tps) ............................................................................................................................. 9 1.2 NTS files (readland.nts) ............................................................................................................................. 9 1.3 Multiple NTS files of single specimens (readmulti.nts) .......................................................................... 10 1.4 Morphologika files (read.morphologika)................................................................................................. 10 1.5 Other text files ......................................................................................................................................... 12 Section 2: Data Preparation: Manipulating landmark data and classifiers ............................................... 14 2.1 Data object formats .................................................................................................................................. 14 3D array (p x k x n)....................................................................................................................................... 14 2D array (n x [p x k]) ................................................................................................................................... 14 2.2 Converting a 2D array into a 3D array (arrayspecs) ................................................................................ 14 2.3 Converting a 3D array into a 2D array (two.d.array) .............................................................................. 15 2.4 Making a factor: group variables ............................................................................................................. 15 2.5 Estimating missing landmarks (estimate.missing) .................................................................................. 16 2.6 Rotate a subset of 2D landmarks to common articulation angle (fixed.angle) ........................................ 18 Section 3a: Generalized Procrustes Analysis ............................................................................................ 19 3.1 Generalized Procrustes Analysis (gpagen) .............................................................................................. 19 3.2 Generalized Procrustes Analysis with Bilateral Symmetry Analysis (bilat.symmetry) .......................... 23 Section 3b: Data Analysis ........................................................................................................................ 26 3.3 Covariation methods: Procrustes ANOVA/regression for shape data (procD.lm) .................................. 26 3.4 Covariation methods: Pairwise Group Comparisons (pairwiseD.test) .................................................... 27 3.5 Covariation methods: Two-block partial least squares analysis for shape data (two.b.pls) .................... 28 3.6 Morphological Integration methods: Quantify morphological integration between two modules
(morphol.integr) ................................................................................................................................................ 29 3.7 Morphological Integration methods: Compare modular signal to alternative landmark subsets
(compare.modular.partitions) ........................................................................................................................... 32 define.modules .............................................................................................................................................. 32 3.8 Phylogenetic Comparative methods: Assessing phylogenetic signal in morphometric data (physignal) 33 3.9 Phylogenetic Comparative methods: Quantify phylogenetic morphological integration between two sets
of variables (phylo.pls) ..................................................................................................................................... 36 3.10 Phylogenetic Comparative methods: Comparing rates of shape evolution on phylogenies
(compare.evol.rates) ......................................................................................................................................... 37 3.11 Calculate morphological disparity for one or more groups (morphol.disparity) ................................... 38 3.12 Quantify and compare shape change trajectories (trajectory.analysis) ................................................. 39 Section 4: Visualization ........................................................................................................................... 42 4.1 Principal Components Analysis (plotTangentSpace) .............................................................................. 42 mshape .......................................................................................................................................................... 44 4.2 Plot allometric patterns in landmark data (plotAllometry) ...................................................................... 44 4.3 Plot phylogenetic tree and specimens in tangent space (plotGMPhyloMorphoSpace) ........................... 47 2
4.4 Create a mesh3d object warped to the mean shape (warpRefMesh) ....................................................... 47 findMeanSpec ............................................................................................................................................... 49 4.5 Plot landmark coordinates for all specimens (plotAllSpecimens) ........................................................... 49 4.6 Plot shape differences between a reference and target specimen (plotRefToTarget).............................. 50 4.7 Plot 3D specimen, fixed landmarks and surface semilandmarks (plotspec) ........................................... 55 Section 5: Data Collection (Digitizing) ...................................................................................................... 56 5.1 2D data collection (digitize2d) ................................................................................................................ 56 5.2 Define sliding semilandmarks in 2D (define.sliders.2d) ......................................................................... 57 5.3 Importing 3D surface files (read.ply) ...................................................................................................... 59 5.4 3D data collection: landmarks (digit.fixed) ............................................................................................. 60 5.5 3D data collection: landmarks and semilandmarks ................................................................................. 62 buildtemplate ................................................................................................................................................ 62 editTemplate ................................................................................................................................................. 65 digitsurface ................................................................................................................................................... 66 5.6 Define sliding semilandmarks in 3D (define.sliders.3d) ......................................................................... 67 Frequently Asked Questions ................................................................................................................... 69 References .............................................................................................................................................. 70 3
Introduction What is geomorph? Geomorph is a freely available software package for geometric morphometric analyses of two- and threedimensional landmark (shape) data in the R statistical computing environment. It can be installed from the
Comprehensive R Archive Network, CRAN http://cran.r-project.org/web/packages/geomorph/.
How to cite: When using geomorph in publications, please cite the software with version and the publication.
Type in R console
> citation(package="geomorph")
Adams, D.C., E. Otarola-Castillo and E. Sherratt. 2014 geomorph: Software for geometric morphometric analyses. R package
version 2.0. http://cran.r-project.org/web/ packages/geomorph/index.html.
Adams, D.C., and E. Otarola-Castillo. 2013. geomorph: an R package for the collection and analysis of geometric morphometric
shape data. Methods in Ecology and Evolution. 4:393-399.
How to use this manual This manual is not meant to be exhaustive – the benefit of working within the R environment is its flexibility
and infinite possibilities. Instead, the manual presents the functions in geomorph and how they can be used
together to perform analyses to address a variety of questions in Biology, Anthropology, Paleontology,
Archaeology, Medicine etc. This help guide is structured according to the pipeline outlined in Figure 1,
which is based on a general workflow for morphometric analysis.
Figure 1 Overview of the morphometric analysis process. In blue are the steps performed in R and geomorph, and those in orange are
done outside of R and imported in.
In section 1, we go over how to import data files of (raw) landmark coordinates digitized elsewhere, e.g.,
using software such as ImageJ or tpsDig for 2D data, or IDAV Landmark editor, AMIRA, Microscribe for
3D data (note that data collection – digitizing landmarks – can also be done in geomorph, and is outlined in
section 5).
4
In section 2, we demonstrate some techniques and functions for preparing and manipulating imported
datasets, such as adding grouping variables and estimating missing data, and adjusting articulated datasets
(2D only). Note that some functions described in this section can also be used on Procrustes coordinate data,
but are presented here because they are important steps to learn familiarize the user with the R environment.
In section 3a, the raw data are taken through the morphometric-specific step of alignment using a
generalized Procrustes superimposition, which is imperative for raw coordinate data. In section 3b, the
statistical analysis functions are presented in order by type of analysis (Table 1). In section 4, we describe
how to plot and visualize the data analysis results, including shape deformation graphs and ordination plots
(e.g., PCA).
Table 1 Functions in geomorph
Data Input
read.morphologika
readland.nts
readland.tps
readmulti.nts
Data Manipulation &
Preparation
arrayspecs
define.modules
estimate.missing
findMeanSpec
mshape
two.d.array
writeland.tps
fixed.angle
Data Analysis
Visualization
Dataset
Digitizing
gpagen
bilat.symmetry
procD.lm
pairwiseD.test
two.b.pls
compare.modular.partitions
morphol.integr
phylo.pls
physignal
compare.evol.rates
morphol.disparity
trajectory.analysis
plotAllometry
plotAllSpecimens
plotGMPhyloMorphoSpace
plotRefToTarget
plotspec
plotTangentSpace
warpRefMesh
hummingbirds
mosquito
motionpaths
plethodon
plethspecies
plethShapeFood
ratland
scallopPLY
scallops
buildtemplate
digit.fixed
digitize2d
digitsurface
editTemplate
read.ply
define.sliders.2d
define.sliders.3d
Section 5 details the functions that can be used to generate coordinate data from 2D images and 3D surface
files (i.e., an ASCII .ply). And finally, at the end there are some frequently asked questions and their
solutions.
Throughout this manual, we will use the following abbreviations as is conventional in morphometrics and R:
n
number of specimens/individuals
p
number of landmarks
k
number of dimensions
#
a comment, in R this is text that is ignored (not run)
...
data not shown
>
denotes the line of code to be written in (the ‘>’ is not typed)
dim() a function
code code to be written into the R console
[1]
in a code example at the start of a line, a number in brackets denotes the first element of the
output and is not intended to be typed
Briefly understanding functions; below is a geomorph function annotated by color:
plotAllometry(A, sz, groups = NULL, method = c("CAC", "RegScore",
"PredLine"), warpgrids = TRUE, iter = 99, label = FALSE, mesh = NULL,
verbose = FALSE)
the function name | an object, usually data, sometimes a file name | a multipart option, requires choice of one
of the presented values | a logical option that requires a TRUE or FALSE input | option that is NULL by
default, but requires either an object or a value | option that requires a value
Usually only the objects are necessary to run a function, as it will use the defaults for the options (which are
presented in the function as above, and under “usage” in the R help pages). Always read the help pages and
check the examples for usage. Order does not matter as long as the option is written in full, e.g., A= mydata.
But “ ” are important, e.g., method = “RegScore”.
5
Workflows for common analyses Below are some pathways to perform common analyses in geomorph. This is not an exhaustive list, but
provides a reference for users familiar with other morphometric software to navigate the functions. In red the
type of question or analysis is presented, and in blue the specific geomorph functions in sequence.
6
Example datasets in geomorph For sections 2 through 4, many of the examples will be using data included with geomorph. There are nine
datasets: plethodon, scallops, hummingbirds, mosquito, ratland, plethspecies, plethShapeFood, motionpaths
and scallopPLY. To load type,
> data(plethodon)
It is advised to run and examine these example datasets before performing own analyses in order to
understand how a function and its options work, and how one’s datasets should be formatted.
> plethodon
$land
, , 1
[1,]
[2,]
[3,]
[4,]
...
$links
[,1]
8.89372
9.26840
5.56104
1.87340
[,2]
53.77644
52.77072
54.21028
52.75100
[,1] [,2]
[1,]
4
5
[2,]
3
5
[3,]
2
4
[4,]
1
2
[5,]
1
3
[6,]
6
7
...
$species
[1] Jord Jord Jord Jord Jord
[18] Teyah Teyah Teyah Jord Jord
[35] Teyah Teyah Teyah Teyah Teyah
Levels: Jord Teyah
$site
[1] Symp Symp Symp Symp Symp Symp
Allo
[22] Allo Allo Allo Allo Allo Allo
Levels: Allo Symp
Jord Jord
Jord Jord
Teyah
Jord
Jord
Jord
Jord
Jord
Jord
Teyah Teyah Teyah Teyah Teyah Teyah Teyah
Jord Jord Jord Teyah Teyah Teyah Teyah
Symp Symp Symp Symp Symp Symp Symp Symp Symp Symp Symp Symp Symp Symp
Allo Allo Allo Allo Allo Allo Allo Allo Allo Allo Allo Allo Allo
The dataset above, plethodon, is a list containing four components: the coordinate data (plethodon$land),
the wirelink addresses for plotting (plethodon$links), and two sets of grouping variables as factors
(plethodon$species, plethodon$site).
> class(plethodon$site)
[1] "factor"
> class(plethodon$land)
[1] "array"
> class(plethodon$links)
[1] "matrix"
Many of the novice user problems when using geomorph and R stem from having the object input in the
wrong format. Here are some useful base functions in R to help understand formatting of one’s data:
class() # Object Classes
attributes() # Object Attribute Lists
dim() # Dimensions of an Object
nrow() ; ncol() # The Number of Rows/Columns of a 2D array
dimnames() # Dimnames of an Object
names() # The Names of an Object
rownames() ; colnames() # Row and Column Names
7
is.numeric() # very useful to know if the data are numeric or not – dataframes often
do not preserved the data as numeric even though they are numbers
Geomorph primarily has data stored in a 2D array or 3D array (matrix and array respectively) (see section
2.1), grouping variables are vectors and factors, and outputs of functions may be lists. For more information
about the object classes in R see http://www.statmethods.net/input/datatypes.html
Permutation tests Many of the function in geomorph test for statistical significance using a permutation procedure (e.g., Good
2000). A randomization test takes the original data, shuffles and resamples, calculates the test statistic and
compares this to the original. This is repeated for a number of iterations, creating a distribution of random
tests statistics in which the original can be evaluated. The proportion of random samples that provide a better
fit to the data than the original provides the P-value. Therefore the number of decimal places for the P-value
is correlated to the number of iterations. When deciding how many iterations to use more is better. However
there is a point where it is time consuming and not helpful (Adams and Anthony 1996). The default in
geomorph is 999, up to 10,000 is reasonable. The examples in this manual and on the package help files are
usually very low to make them fast to run, and it is not recommended to run the function at these small
iterations for one’s own data.
Finally, this manual only covers geomorph functions. It is recommended that users look to some “getting
started with R” resources, such as Quick-R (http://www.statmethods.net/), and the R Introduction manual on
(http://www.r-project.org/), and various Springer eBooks in the series ‘Use R!’. Also highly recommended is
J. Claude’s book Morphometric with R (2008).
8
Section 1: Data Input: Importing landmark data 1.1 TPS files (readland.tps) Function:
readland.tps(file, specID = c("None", "ID", "imageID"))
This function reads a *.tps file containing two- or three-dimensional landmark coordinates for a set of
specimens. Tps files are text files in one of the standard formats for geometric morphometrics (see Rohlf
2010). Two-dimensional landmarks coordinates are designated by the identifier "LM=", while threedimensional data are designated by "LM3=". Landmark coordinates are multiplied by their scale factor if this
is provided for all specimens. If one or more specimens are missing the scale factor (there is no line
“SCALE=”), landmarks are treated in their original units. The name of the specimen can be given in the tps
file by “ID=” (use specID=”ID”) or “IMAGE=” (use specID= “imageID”), otherwise the function defaults
to specID= “None”.
ratland.tps
LM=8
-0.45 -0.475
-0.59 -0.28
-0.515 -0.12
-0.33 0
0 0
0.145 -0.395
-0.045 -0.42
-0.26 -0.465
ID = specimen103N
LM=8
...
> mydata <- readland.tps("ratland.tps", specID = "ID")
[1] "Not all specimens have scale. Using scale = 1.0"
> mydata[,,1]
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
[,1]
-0.450
-0.590
-0.515
-0.330
0.000
0.145
-0.045
-0.260
[,2]
-0.475
-0.280
-0.120
0.000
0.000
-0.395
-0.420
-0.465
In this case, there is no scale given in the tps file, so the command warns that the data are treated in their
original units. The function returns a 3D array containing the coordinate data, and if provided in the file, the
names of the specimens (dimnames(mydata)[[3]]).
1.2 NTS files (readland.nts) Function:
readland.nts(file)
Function reads *.nts file containing a matrix of two- or three-dimensional landmark coordinates for a set of
specimens. NTS files are text files in one of the standard formats for geometric morphometrics (see Rohlf
2012). The parameter line contains 5 or 6 elements, and must begin with a "1" to designate a rectangular
matrix. The second and third values designate how many specimens (n) and how many total variables (p x k)
9
are in the data matrix. The fourth value is a "0" if the data matrix is complete and a "1" if there are missing
values. If missing values are present, the '1' is followed by the arbitrary numeric code used to represent
missing values (e.g., -999). These values will be replaced with "NA" in the output array. The final value of
the parameter line denotes the dimensionality of the landmarks (2,3) and begins with "DIM=". If specimen
and variable labels are included, these are designated placing an "L" immediately following the specimen or
variable values in the parameter file. The labels then precede the data matrix. Here there are n = 44 and p*k =
50 (25 2D landmarks).
rats.nts
" rats data, 164 rats, 8 landmarks in 2 dimensions
1 164 16 0 dim=2
-.450 -.475 -.590 -.280 -.515 -.120 -.330 0 0 0 .145
-.530 -.555 -.685 -.320 -.625 -.120 -.400 0 0 0 .230
-.560 -.570 -.700 -.335 -.670 -.120 -.425 0 0 0 .300
-.590 -.580 -.745 -.355 -.700 -.100 -.435 0 0 0 .330
-.650 -.580 -.800 -.340 -.715 -.090 -.450 0 0 0 .360
...
-.395
-.425
-.440
-.445
-.445
-.045 -.420 -.260 -.465
-.005 -.480 -.265 -.525
.015 -.495 -.270 -.540
.030 -.505 -.285 -.565
.040 -.515 -.300 -.580
> mydata <- readland.nts("rats.nts")
> mydata[,,1]
[1,]
[2,]
[3,]
[4,]
...
[,1]
[,2]
-0.450 -0.475
-0.590 -0.280
-0.515 -0.120
-0.330 0.000
The function returns a 3D array containing the coordinate data, and if provided in the file, the names of the
specimens (dimnames(mydata)[[3]]).
Function is for *.nts file containing landmark coordinates for multiple specimens. Note that *.dta files in the
nts format written by Landmark Editor http://graphics.idav.ucdavis.edu/research/projects/EvoMorph, and
*.nts files written by Stratovan Checkpoint http://www.stratovan.com/ have incorrect header notation; every
header is 1 n p-x-k 1 9999 Dim=3, rather than 1 n p-x-k 0 Dim=3, which denotes that missing data is in the
file even when it is not. NAs will be introduced unless the header is manually.
1.3 Multiple NTS files of single specimens (readmulti.nts) Function:
readmulti.nts(file)
This function reads a list containing the names of multiple *.nts files, where each contains the landmark
coordinates for a single specimen. For these files, the number of variables (columns) of the data matrix will
equal the number of dimensions of the landmark data (k=2 or 3). When the function is called a dialog box is
opened, from which the user may select multiple *.nts files. These are then read and concatenated into a
single matrix for all specimens.
> filelist <- list.files(pattern = ".nts")
> mydata <- readmulti.nts(filelist)
The function returns a 3D array containing the coordinate data, and the names of the specimens
(dimnames(mydata)[[3]]) extracted from the file names.
1.4 Morphologika files (read.morphologika) Function:
read,morphologika(file)
10
This function reads a *.txt file in the Morphologika format containing two- or three-dimensional landmark
coordinates. Morphologika files are text files in one of the standard formats for geometric morphometrics
(see O'Higgins and Jones 1998), see http://sites.google.com/site/hymsfme/resources. If the headers "[labels]"
and "[labelvalues]" are present in the file, then a data matrix containing all individual specimen information
is returned. If the header "[wireframe]" is present, then a matrix of the landmark addresses for the wireframe
is returned.
morphologikaexample.txt
[individuals]
15
[landmarks]
31
[Dimensions]
3
[names]
Specimen 1
Specimen 2
Specimen 3
...
Specimen 15
[labels]
Sex
[labelvalues]
Female
Female
...
Female
[rawpoints]
'#1
16.01 24.17 11.18
15 24.86 11.16
14.96 25.54 11.52
16.26 24.36 11.48
15.89 26.61 11.83
17.16 25.33 12.35
18.22 23.65 11.12
...
> mydata <- read.morphologika("morphologikaexample.txt")
> mydata$coords[,,1]
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
...
[,1]
16.01
15.00
14.96
16.26
15.89
17.16
[,2]
24.17
24.86
25.54
24.36
26.61
25.33
[,3]
11.18
11.16
11.52
11.48
11.83
12.35
> mydata$labels
Specimen
Specimen
Specimen
Specimen
Specimen
Specimen
...
1
2
3
4
5
6
Sex
"Female"
"Female"
"Female"
"Female"
"Male"
"Male"
The function returns a 3D array containing the coordinate data, and if provided, the names of the specimens
(dimnames(mydata)[[3]]). If other optional headers are present in the file (e.g., "[labels]" or "[wireframe]")
function returns a list containing the 3D array of coordinates ($coords), and a data matrix of the data from
11
labels ($labels) and/or the landmark addresses demoting the wireframe ($wireframe) – which can be
passed to see plotRefToTarget() option 'links').
1.5 Other text files Base Functions:
read.table(file)
read.csv(file)
Using base read functions in R, one can read in data by many other ways. Here are two examples. These
examples use data arrangement function arrayspecs() (see section 2.1 for details) and creates an object in
the same way as the previous functions.
For a set of files (file1.txt, file2.txt, file3.txt...) each containing the landmark coordinates of a single
specimen
file1.txt
16.01 24.17 11.18
15 24.86 11.16
14.96 25.54 11.52
16.26 24.36 11.48
15.89 26.61 11.83
17.16 25.33 12.35
18.22 23.65 11.12
...
> filelist <- list.files(pattern = ".txt") # makes a list of all .txt files in
working directory
> names <- gsub (".txt", "", filelist) # extracts names of specimens from the file
name
> coords = NULL # make an empty object
> for (i in 1:length(filelist)){
tmp <- as.matrix(read.table(filelist[i]))
coords <- rbind(coords, tmp)
}
> coords <- arrayspecs(coords, p, k)
> dimnames(coords)[[3]] <- names
For a files each containing the landmark coordinates of a set of specimens, where each row is a specimen,
and coordinate data arranged in columns x1, y1, x2, y2… etc., and the first column is the ID of the specimens,
e.g., from a data file exported from MorphoJ.
coordinatedata.txt
ID X1 Y1 X2 Y2 X3 Y3 ...
specimen1 0.595 0.1679 0.2232 0.5028 1.292 0.4237 0.51 ...
specimen2 0.0038 1.3925 0.7966 0.4132 0.1006 0.8483 ...
specimen3 0.6249 0.4515 0.3576 1.3262 0.9114 0.3611 ...
...
>
>
>
>
tmp <- as.matrix(read.table("coordinatedata.txt", header=T))
names <- tmp[,1]
coords <- arrayspecs(tmp[,2:ncol(tmp)], p, k)
dimnames(coords)[[3]] <- names
If it is a data file exported from MorphoJ, and the centroid size and/or classifiers are included as columns
before the data, follow the example above and replace the “2” on coords<- line with the column number the
coordinate data start on. Then one can make a second matrix of the classifier data by:
12
> classifiers <- tmp[,1:x] # taking only the columns containing classifiers, where x
is the number of the last column before the coordinate data.
13
Section 2: Data Preparation: Manipulating landmark data and classifiers 2.1 Data object formats Landmark data in geomorph can be found as objects in two formats: a 2D array (matrix; Figure 2A) or a 3D
array (Figure 2B). These data formats follow the convention in other morphometric packages (e.g., shapes,
Morpho) and in J.Claude’s book Morphometrics in R (2008).
B(
A(
X1# Y1# Z1# X2# Y2# Z2# X3# Y3# ...# Xp#
#
Specimen#1#
Specimen#2#
Specimen(n(
Specimen#3#
...#
Specimen#n#
Specimen(3(
Specimen(2(
Specimen(1(
Figure 2 Schematic of a 2D array (A) and a 3D array (B). This example shows 3D landmark coordinate data, but the same format would
be used for 2D coordinates
3D array (p x k x n) An array with three dimensions, i.e., number of rows (p), number of columns (k) and number of “sheets” (n).
Imagine a 3D array like a stack of filing cards. Data in this format are needed for most geomorph analysis
functions. If one has inputted data using
readland.nts(), readmulti.nts(), readland.tps(), read.morphologika()
Then the data will be a 3D array object. Check by typing
> dim(mydata)
[1] 15 3 40
If dim() gives three numbers, it is a 3D array. Here mydata has p=15, k=3, n=40.
If dim() gives two number, it is a 2D array (a matrix).
2D array (n x [p x k]) An array (matrix) with two dimensions, i.e., number of rows (n) and number of columns (p*k). From the
example above
> dim(mydata)
[1] 40 30
Data in this format are needed for the following geomorph analysis functions (see section 3):
procD.lm(), trajectory.analysis(), pairwise.D.test()
2.2 Converting a 2D array into a 3D array (arrayspecs) Function:
arrayspecs(A, p, k)
This function converts a matrix of landmark coordinates into a 3D array (p x k x n), which is the required
input format for many functions in geomorph. The input matrix can be arranged such that the coordinates of
14
each landmark are found on a separate row, or that each row contains all landmark coordinates for a single
specimen.
> A <- arrayspecs(mydata, p, k) # where mydata is a 2D array
> A[,,1] # look at just the first specimen
, , 1
[,1]
[,2]
[1,] 8.89372 53.77644
[2,] 9.26840 52.77072
[3,] 5.56104 54.21028
[4,] 1.87340 52.75100
[5,] 1.28180 53.18484
[6,] 1.24236 53.32288
[7,] 0.84796 54.70328
[8,] 3.35240 55.76816
[9,] 6.29068 55.70900
[10,] 8.87400 55.25544
[11,] 10.74740 55.43292
[12,] 14.39560 52.75100
2.3 Converting a 3D array into a 2D array (two.d.array) Function:
two.d.array(A)
This function converts a 3D array (p x k x n) of landmark coordinates into a 2D array (n x [p x k]). The latter
format of the shape data is useful for performing subsequent statistical analyses in R (e.g., PCA, MANOVA,
PLS, etc.). Row labels are preserved if included in the original array.
> a <- two.d.array(mydata)
> a
[,1]
[1,] 8.893720
[2,] 8.679762
[3,] 9.805328
[4,] 9.637164
[5,] 11.035692
...
# where mydata is a 3D array
[,2]
[,3]
[,4]
[,5]
53.77644 9.268400 52.77072 5.561040
54.57819 8.935628 53.83027 5.451914
56.06903 10.137712 55.27961 6.647680
58.03294 9.952104 56.77318 6.109836
58.75009 11.335110 57.85184 8.255382
[,6]
54.21028
54.65691
55.73664
57.94896
58.92119
[,7]
1.873400
1.987882
3.448484
2.645496
4.555431
[,8]
52.75100
52.68871
53.86698
55.89135
57.46687
[,9]
1.281800
1.515514
3.012230
2.015616
3.956595
[,10]
53.18484
53.02331
54.34478
56.62621
58.08709
2.4 Making a factor: group variables Many analyses will require a grouping variable (a classifier) for the data. For small datasets, this can be made
easily within R:
> group <- factor(c(0,0,1,0,0,1,1,0,0)) # specimens assigned in order to group 0 or
1
> names(group) <- dimnames(mydata)[[3]] # assign specimen names from 3D array of
data to the group classifier
> group
[1] 0 0 1 0 0 1 1 0 0
Levels: 0 1
15
If the data have many specimens or many different groups, it may be easier to make a table in excel, save as
a .csv file and import using read.csv().
classifier.csv
ID#
Species#
Habitat#
specimen1#
A.species#
dry#
specimen2#
A.notherspecies#
wet#
specimen3#
A.species#
dry#
specimen4#
A.species#
dry#
specimen5#
A.species#
wet#
specimen6#
A.notherspecies#
wet#
specimen7#
A.notherspecies#
wet#
specimen8#
A.notherspecies#
wet#
specimen9#
A.notherspecies#
dry#
...#
...#
...#
> classifier <- read.csv("classifier.csv", header=T, row.names=1)
> is.factor(classifier$Habitat) # check that it is a factor
[1] TRUE
> classifier$Habitat
[1] dry wet dry dry wet wet wet wet dry ...
Levels: dry wet
2.5 Estimating missing landmarks (estimate.missing) All analysis and plotting functions in geomorph require a full complement of landmark coordinates.
Either the missing values are estimated, or subsequent analyses are performed on a subset dataset excluding
specimens with missing values. Below is the function to estimate missing data, followed by steps of how to
just exclude specimens with missing values.
Function:
estimate.missing(A, method = c("TPS", "Reg"))
Arguments:
A
method
A 3D array (p x k x n) containing landmark coordinates for a set of specimens
Method for estimating missing landmark locations
The function estimates the locations of missing landmarks for incomplete specimens in a set of landmark
configurations, where missing landmarks in the incomplete specimens are designated by NA in place of the
x,y,z coordinates. Two distinct approaches are implemented.
The first approach (method="TPS") uses the thin-plate spline to interpolate landmarks on a reference
specimen to estimate the locations of missing landmarks on a target specimen. Here, a reference specimen is
obtained from the set of specimens for which all landmarks are present, Next, each incomplete specimen is
aligned to the reference using the set of landmarks common to both. Finally, the thin-plate spline is used to
estimate the locations of the missing landmarks in the target specimen (Gunz et al. 2009).
The second approach (method="Reg") is multivariate regression. Here each landmark with missing values is
regressed on all other landmarks for the set of complete specimens, and the missing landmark values are then
predicted by this linear regression model. Because the number of variables can exceed the number of
specimens, the regression is implemented on scores along the first set of PLS axes for the complete and
incomplete blocks of landmarks (see Gunz et al. 2009).
16
One can also exploit bilateral symmetry to estimate the locations of missing landmarks. Several possibilities
exist for implementing this approach (see Gunz et al. 2009). Example R code for one implementation is
found in Claude (2008).
Missing landmarks in a target specimen are designated by NA in place of the x,y,z coordinates. To make this
so,
> any(is.na(mydata) # check if there are NAs in the data
FALSE # if false then,
> mydata[which(mydata == -999)] <- NA #change missing values from “-999” to NAs
Here is an example using Plethodon dataset,
>
>
>
>
>
>
>
data(plethodon)
plethland<-plethodon$land
plethland[2,,2]<-plethland[6,,2]<-NA #create missing landmarks
plethland[2,,5]<-plethland[6,,5]<-NA
plethland[2,,10]<-NA
estimate.missing(plethland,method="TPS")
estimate.missing(plethland,method="Reg")
The function returns a 3D array with the missing landmarks estimated.
Instead of estimating missing, an alternative is to proceed with the specimens for which data are missing
excluded. For example to make a dataset of only the complete specimens (starting with the dataset as 2D
array), two ways are possible:
> mydata
[,1]
[,2]
[,3]
[1,] 8.893720 53.77644 9.268400
[2,] 8.679762 54.57819 8.935628
[3,] 9.805328 56.06903 NA NA
[4,] 9.637164 58.03294 9.952104
[5,] NA NA 11.335110 57.85184
[6,] 7.946625 55.71114 8.476400
[7,] 8.849841 58.66961 9.396387
[8,] 9.331504 56.36904 10.154872
[,4]
52.77072
53.83027
56.77318
54.82112
57.82877
55.31344
> newdata <- mydata[complete.cases(mydata),] # keep only specimens with complete
data
> # OR
> newdata <- na.omit(mydata) # use only specimens without NAs
> newdata
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[,1]
8.893720
8.679762
9.637164
7.946625
8.849841
9.331504
[,2]
[,3]
53.77644 9.268400
54.57819 8.935628
58.03294 9.952104
55.71114 8.476400
58.66961 9.396387
56.36904 10.154872
[,4]
52.77072
53.83027
56.77318
54.82112
57.82877
55.31344
These same functions can be used to make a dataset of only the landmarks in all specimens, by inputting the
matrix mydata in transpose, e.g., t(mydata). Note that these methods will re-label the specimen or landmark
numbers.
17
2.6 Rotate a subset of 2D landmarks to common articulation angle (fixed.angle) A function for rotating a subset of landmarks so that the articulation angle between subsets is constant.
Presently, the function is only implemented for two-dimensional landmark data.
Function:
fixed.angle(A, art.pt = NULL, angle.pts = NULL, rot.pts = NULL,
angle = 0, degrees = FALSE)
Arguments:
A
art.pt
angle.pts
rot.pts
angle
degrees
A 3D array (p x k x n) containing landmark coordinates for a set of specimens
A number specifying which landmark is the articulation point between the two landmark subsets
A vector containing numbers specifying which two points used to define the angle (one per subset)
A vector containing numbers specifying which landmarks are in the subset to be rotated
An optional value specifying the additional amount by which the rotation should be augmented (in radians)
A logical value specifying whether the additional rotation angle is expressed in degrees or radians (radians is default)
This function standardizes the angle between two subsets of landmarks for a set of specimens. The approach
assumes a simple hinge-point articulation between the two subsets, and rotates all specimens such that the
angle between landmark subsets is equal across specimens (see Adams 1999). As a default, the mean angle is
used, though the user may specify an additional amount by which this may be augmented.
Example using Plethodon. Articulation point is landmark 1, rotate mandibular landmarks (2-5) relative to
cranium
> data(plethspecies)
> fixed.angle(plethspecies$land,art.pt=1,angle.pts=c(5,6),rot.pts=c(2,3,4,5))
Before
After
Function returns a 3D array containing the newly rotated data.
18
Section 3a: Generalized Procrustes Analysis 3.1
Generalized Procrustes Analysis (gpagen) Generalized Procrustes Analysis (GPA: Gower 1975; Rohlf and Slice 1990) is the primary means by which
shape variables are obtained from landmark data (for a general overview of geometric morphometrics see
Bookstein 1991; Rohlf and Marcus 1993; Adams et al. 2004; Mitteroecker and Gunz 2009; Zelditch et al.
2012; Adams et al. 2013). GPA translates all specimens to the origin, scales them to unit-centroid size, and
optimally rotates them (using a least-squares criterion) until the coordinates of corresponding points align as
closely as possible. The resulting aligned Procrustes coordinates represent the shape of each specimen, and
are found in a curved space related to Kendall's shape space (Kendall 1984). Typically, these are projected
into a linear tangent space yielding Kendall's tangent space coordinates (Dryden and Mardia 1993; Rohlf
1999), which are used for subsequent multivariate analyses. Additionally, any semilandmarks on curves and
are slid along their tangent directions or tangent planes during the superimposition (see Bookstein 1997;
Gunz et al. 2005). Presently, two implementations are possible: 1) the locations of semilandmarks can be
optimized by minimizing the bending energy between the reference and target specimen (Bookstein 1997), or
by minimizing the Procrustes distance between the two (Rohlf 2010).
The first step in any geometric morphometric analysis is to perform a superimposition of the raw
coordinate data (read in using functions in section 1).
Function:
gpagen(A, Proj = TRUE, ProcD = TRUE, ShowPlot = TRUE, curves = NULL,
surfaces = NULL, pointscale = 1)
Arguments:
A
Proj
ProcD
curves
pointscale
surfaces
ShowPlot
A 3D array (p x k x n) containing landmark coordinates for a set of specimens
A logical value indicating whether or not the aligned Procrustes residuals should be projected into tangent space
A logical value indicating whether or not Procrustes distance should be used as the criterion for optimizing the
positions of semilandmarks
An optional matrix defining which landmarks should be treated as semilandmarks on boundary curves, and which
landmarks specify the tangent directions for their sliding (see define.sliders.2d() or
define.sliders.3d())
An optional value defining the size of the points for all specimens
An optional vector defining which landmarks should be treated as semilandmarks on surfaces
A logical value indicating whether or not a plot of Procrustes residuals should be displayed
The function performs a Generalized Procrustes Analysis (GPA) on two-dimensional or three-dimensional
landmark coordinates. The analysis can be performed on fixed landmark points, semilandmarks on curves,
semilandmarks on surfaces, or any combination. To include semilandmarks on curves, one must specify a
matrix defining which landmarks are to be treated as semilandmarks using the "curves=" option (this matrix
can be made using define.sliders.2d() or define.sliders.3d(), see section 5). Likewise, to include
semilandmarks on surfaces, one must specify a vector listing which landmarks are to be treated as surface
semilandmarks using the "surfaces=" option. The "ProcD=TRUE" option will slide the semilandmarks along
their tangent directions using the Procrustes distance criterion, while "ProcD=FALSE" will slide the
semilandmarks based on minimizing bending energy. The aligned Procrustes residuals can be projected into
tangent space using the "Proj=TRUE" option. NOTE: Large datasets may exceed the memory limitations of R.
If the curve semilandmarks were defined by define.sliders.2d() or define.sliders.3d(), or the
surface semilandmarks digitized with buildtemplate() and digitizesurface() the .csv files these
functions make must be first read in as follows:
> curves <- as.matrix(read.csv("curveslide.csv", header=T))
> sliders <- as.matrix(read.csv("surfslide.csv", header=T))
19
Example using fixed points only:
> data(plethodon)
> Y <- gpagen(plethodon$land)
> Y
$coords
, , 1
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
...
[,1]
0.18705239
0.21178322
-0.03236886
-0.27528209
-0.31422144
-0.31685794
-0.34273719
[,2]
-0.023704219
-0.089958949
0.004834644
-0.091255419
-0.062671167
-0.053529677
0.037340225
$Csize
[1] 15.226920 14.718881 14.009548 15.750465 13.935179 13.506818 13.593187 16.020895 15.641445 16.380327
[11] 19.083462 16.930622 18.426860 9.581975 19.688119 18.233979 11.290599 18.980799 17.009811 17.067253
...
The function returns a list containing the Procrustes coordinates ($coords), the centroid sizes ($Csize) and,
if ShowPlot=T a graph of the specimens (grey) around the mean (black) shape:
Example using fixed points and semilandmarks on curves:
> data(hummingbirds)
> hummingbirds$curvepts # Matrix defining which points are semilandmarks (middle
column) and in which directions they slide (columns 1 vs. 3)
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
...
before slide after
1
11
12
11
12
13
13
14
15
7
15
14
12
13
14
1
16
17
16
17
18
17
18
19
#Using Procrustes Distance for sliding
> Y <- gpagen(hummingbirds$land,curves=hummingbirds$curvepts)
> Y
>
$coords
20
, , 1
[,1]
[,2]
[1,] -0.24134653 -0.035692519
[2,] -0.23007167 -0.034486535
[3,] -0.13498860 -0.007088872
[4,] -0.11202212 -0.002656968
[5,] 0.25340294 0.035622930
...
$Csize
[1] 2538.998 2683.454 2659.923 2715.072 2640.722 2779.321 2547.102 2717.373 2626.064 2870.187 2631.083 ...
> #Using bending energy for sliding
> Y <- gpagen(hummingbirds$land,curves=hummingbirds$curvepts,ProcD=FALSE)
> Y
[,1]
[,2]
[1,] -0.23767890 -0.034789230
[2,] -0.22636001 -0.033583624
[3,] -0.13077781 -0.006039000
[4,] -0.10768707 -0.001583413
[5,] 0.25965630 0.036913166
...
$Csize
[1] 2538.998 2683.454 2659.923 2715.072 2640.722 2779.321 2547.102 2717.373 2626.064 2870.187 2631.083 ...
Example using fixed points, curves and surfaces:
> data(scallops)
> scallops$curvslide # Matrix defining which points are semilandmarks (middle
column) and in which directions they slide (columns 1 vs. 3)
[1,]
[2,]
before slide after
5
6
7
6
7
8
21
[3,]
[4,]
[5,]
[6,]
[7,]
...
7
8
9
10
11
8
9
10
11
12
9
10
11
12
13
> scallops$surfslide # Matrix (1 column) defining which points are semilandmarks to
slide over the surface
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
[9,]
...
[,1]
17
18
19
20
21
22
23
24
25
> #Using Procrustes Distance for sliding
> Y <- gpagen(A=scallops$coorddata, curves=scallops$curvslide,
surfaces=scallops$surfslide)
> Y
$coords
, , ZXkuhn01Lxcoords
[,1]
[,2]
[,3]
[1,] -0.108197649 -0.0539704198 0.068318470
[2,] -0.091387002 -0.0871397342 0.123700784
[3,] -0.006796579 -0.0829786905 0.129445557
[4,] 0.098543611 -0.0764721891 0.137365515
[5,] 0.109359514 -0.0447354705 0.092469791
[6,] 0.166434371 -0.0243779199 0.062752544
[7,] 0.185504252 0.0008896398 0.008507093
[8,] 0.175604912 0.0242133843 -0.051991461
...
$Csize
ZXkuhn01Lxcoords
ZXkuhn03Lxcoords
subn01Lcoords irradiansls13coords
188.08311
189.12024
266.37888
92.95572
colb01Lcoords
154.91506
For 3D data, the Procrustes data are plotted in the rgl window.
For all subsequent analyses, the Procrustes coordinates (Y$coords) should be used.
22
3.2
Generalized Procrustes Analysis with Bilateral Symmetry Analysis (bilat.symmetry) If the data has bilateral symmetry, the first step is to perform a superimposition of the raw coordinate data
taking into account the symmetry. This function also assesses the statistical differences in the symmetric data.
Function:
bilat.symmetry(A, ind = NULL, side = NULL, replicate = NULL,
object.sym = FALSE, land.pairs = NULL, warpgrids = TRUE, mesh = NULL,
verbose = FALSE)
Arguments:
A
ind
side
replicate
object.sym
land.pairs
warpgrids
mesh
verbose
A 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens [for "object.sym=FALSE, A
is of dimension (n x k x 2n)]
A vector containing labels for each individual. For matching symmetry, the matched pairs receive the same label
(replicates also receive the same label).
An optional vector (for matching symmetry) designating which object belongs to which 'side-group'
An optional vector designating which objects belong to which group of replicates
A logical value specifying whether the analysis should proceed based on object symmetry =TRUE or matching
symmetry =FALSE
An optional matrix (for object symmetry) containing numbers for matched pairs of landmarks across the line of
symmetry
A logical value indicating whether deformation grids for directional and fluctuating components of asymmetry
A mesh3d object to be warped to represent shape deformation of the directional and fluctuating components of
asymmetry if warpgrids= TRUE (see warpRefMesh).
A logical value indicating whether the output is basic or verbose
The function quantifies components of shape variation for a set of specimens as described by their patterns of
symmetry and asymmetry. Here, shape variation is decomposed into variation among individuals, variation
among sides (directional asymmetry), and variation due to an individual x side interaction (fluctuating
symmetry). These components are then statistically evaluated using Procrustes ANOVA and Goodall's F tests
(i.e., an isotropic model of shape variation). Methods for both matching symmetry and object symmetry can
be implemented. Matching symmetry is when each object contains mirrored pairs of structures (e.g., right
and left hands) while object symmetry is when a single object is symmetric about a midline (e.g., right and
left sides of human faces). Analytical and computational details concerning the analysis of symmetry in
geometric morphometrics can be found in Mardia et al. (2000) and Klingenberg et al. (2002).
Analyses of symmetry for matched pairs of objects is implemented when object.sym=FALSE. Here, a 3D
array [p x k x 2n] contains the landmark coordinates for all pairs of structures (2 structures for each of n
specimens). Because the two sets of structures are on opposite sides, they represent mirror images, and one
set must be reflected prior to the analysis to allow landmark correspondence. It is assumed that the user has
done this prior to performing the symmetry analysis. Reflecting a set of specimens may be accomplished
by multiplying one coordinate dimension by '-1' for these structures (either the x-, the y-, or the z-dimension).
A vector containing information on individuals and sides must also be supplied. Replicates of each specimen
may also be included in the dataset, and when specified will be used as measurement error (see Klingenberg
and McIntyre 1998).
Analyses of object symmetry is implemented when object.sym=TRUE. Here, a 3D array [p x k x n] contains
the landmark coordinates for all n specimens. To obtain information about asymmetry, the function generates
a second set of objects by reflecting them about one of their coordinate axes. The landmarks across the line
of symmetry are then relabeled to obtain landmark correspondence. The user must supply a list of landmark
pairs. A vector containing information on individuals must also be supplied. Replicates of each specimen
may also be included in the dataset, and when specified will be used as measurement error.
23
Example of matching symmetry:
> data(mosquito)
> bilat.symmetry(mosquito$wingshape,ind=mosquito$ind,side=mosquito$side,
replicate=mosquito$replicate,object.sym=FALSE, verbose=F)
$ANOVA.size
ind
side
ind:side
replicate
Df
9
1
9
20
Sum Sq
4.149667e-09
3.473794e-10
6.956898e-09
1.090984e-08
Mean Sq
F
P
4.610741e-10 0.5964824 0.7733295
3.473794e-10 0.4493978 0.5194505
7.729887e-10 1.4170492 0.2459017
5.454918e-10
NA
NA
$ANOVA.Shape
df
SS.obs
MS F.Goodall
P.param
F
P
ind
288 0.104888121 0.0003641949
0.5 0.45532641 2.6900999 1.110223e-16
side
32 0.003220857 0.0001006518
1.0 0.01398196 0.7434574 8.433084e-01
ind:side
288 0.038990419 0.0001353834
1.0 0.16926004 1.0406766 3.406141e-01
ind:side:replicate 640 0.083258692 0.0001300917
1.0 0.36143160
NA
NA
Function returns the ANOVA table for analysis of symmetry and a graph showing the shape deformations
relating to the symmetric and asymmetric components of shape. When verbose=TRUE, the function returns
the symmetric component of shape variation ($symm.shape) and the asymmetric component of shape
variation ($asymm.shape), to be used in subsequent analyses like Procrustes coordinates.
24
Example of object symmetry:
> data(scallops)
> bilat.symmetry(scallops$coorddata, ind=scallops$ind, object.sym=TRUE,
land.pairs=scallops$land.pairs)
[1] "No specimen names in response matrix. Assuming specimens in same order."
df
SS.obs
MS F.Goodall
P.param
F P
ind
276 0.064029966 2.319926e-04
0.5 0.64430682 8.837028 0
side
62 0.028837521 4.651213e-04
0.5 0.29017994 17.717330 0
ind:side 248 0.006510579 2.625234e-05
1.0 0.06551324
NA NA
25
Section 3b: Data Analysis After the data have been superimposed with gpagen() or bilat.symmetry(), the Procrustes coordinates can
be used in many statistical analyses (this section), ordination methods and visualization methods (Section 4).
Figure 3 Overview of the analysis (blue) and visualization (green) functions in geomorph.
Protip! Throughout the analysis functions, one will see the option verbose=TRUE/FALSE. This is an option to
limit the amount of information returned. FALSE is default, and the basic results are returned, usually the test
statistic and P-value. When TRUE, the function will also return new datasets, and shape deformation
coordinates that can be used in other functions. Look out for this option in many of the functions.
Several functions have the option “mesh=”, which is there for 3D data users that wish to view the shape
deformations as a warped surface mesh. Further details in warpRefMesh().
3.3
Covariation methods: Procrustes ANOVA/regression for shape data (procD.lm) Function performs Procrustes ANOVA with permutation procedures to assess statistical hypotheses
describing patterns of shape variation and covariation for a set of Procrustes-aligned coordinates.
Function:
procD.lm(f1, data = NULL, iter = 999)
Arguments:
f1
data
A formula for the linear model (e.g., y~x1+x2), where y is a two-dimensional array of shape data
An optional value specifying a data frame containing all data (not required)
26
iter
Number of iterations for significance testing
The function quantifies the relative amount of shape variation attributable to one or more factors in a linear
model and assesses this variation via permutation. Data input is specified by a formula (e.g., y~X), where 'y'
specifies the response variables (shape data), and 'X' contains one or more independent variables (discrete or
continuous). The response matrix 'y' must be in the form of a 2D array rather than a 3D array. The names
specified for the independent (x) variables in the formula represent one or more vectors containing
continuous data or factors. It is assumed that the order of the specimens in the shape matrix matches the order
of values in the independent variables.
The function performs statistical assessment of the terms in the model using Procrustes distances among
specimens, rather than explained covariance matrices among variables. With this approach, the sum-ofsquared Procrustes distances are used as a measure of SS (see Goodall 1991). The observed SS are evaluated
through permutation, where the rows of the shape matrix are randomized relative to the design matrix.
Procedurally, Procrustes ANOVA is identical to permutational-MANOVA as used in other fields (Anderson
2001). For several reasons, Procrustes ANOVA is particularly useful for shape data. First, covariance
matrices from GPA-aligned Procrustes coordinates are singular, and thus standard approaches such as
MANOVA cannot be accomplished unless generalized inverses are utilized. This problem is accentuated
when using sliding semilandmarks. Additionally, geometric morphometric datasets often have more variables
than specimens (the 'small N large P' problem). In these cases, distance-based procedures can still be utilized
to assess statistical hypotheses, whereas standard linear models cannot.
Example of a allometric regression, y~x(continuous) :
> data(ratland)
> rat.gpa<-gpagen(ratland)
#GPA-alignment
> procD.lm(two.d.array(rat.gpa$coords)~rat.gpa$Csize,iter=99)
[1] "No specimen names in response matrix. Assuming specimens in same order."
df
SS.obs
MS
F P.val
Rsq
rat.gpa$Csize
1 0.6403531 0.640353114 513.4641 0.01 0.7601649
Total
163 0.8423871 0.005168019
NA
NA
NA
This example tests for the association between centroid size and shape. If it is significant, we suggest
examining the allometric shape variation using plotAllometry().
Example of a MANOVA for Goodall's F test, y~x(discrete):
>
>
>
>
data(plethodon)
Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
y<-two.d.array(Y.gpa$coords)
procD.lm(y~plethodon$species*plethodon$site,iter=99)
[1] "No specimen names in response matrix. Assuming specimens in same order."
df
SS.obs
MS
F P.val
Rsq
plethodon$species
1 0.02925784 0.029257838 14.54367 0.01 0.1485624
plethodon$site
1 0.06437484 0.064374844 31.99985 0.01 0.3268759
plethodon$species:plethodon$site 1 0.03088502 0.030885020 15.35252 0.01 0.1568247
Total
39 0.19693973 0.005049737
NA
NA
NA
Visualization of the shape changes can be done with plotRefToTarget().
3.4
Covariation methods: Pairwise Group Comparisons (pairwiseD.test) 27
The function is designed as a post-hoc test to Procrustes ANOVA (3.3), where the latter has identified
significant shape variation explained by a grouping factor.
Function:
pairwiseD.test(y, x, iter = 999)
Arguments:
y
x
iter
A two-dimensional array of shape data
A factor defining groups
Number of iterations for permutation test
The function performs pairwise comparisons to identify shape among groups. The function takes as input the
shape data (y), and a grouping factor (x). It then estimates the Euclidean distances among group means,
which are used as test values. These are then statistically evaluated through permutation, where the rows of
the shape matrix are randomized relative to the grouping variable. The input for the shape data (y) must be in
the form of a 2D array rather than a 3D array.
>
>
>
>
>
data(plethodon)
Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
y<-two.d.array(Y.gpa$coords)
procD.lm(y~plethodon$species,iter=99) # Procrustes ANOVA
pairwiseD.test(y,plethodon$species,iter=99)
$Dist.obs
Jord
Teyah
Jord 0.00000000 0.05409051
Teyah 0.05409051 0.00000000
$Prob.Dist
Jord Teyah
Jord 1.00 0.01
Teyah 0.01 1.00
Function returns a list with a matrix of the Euclidean distances among groups ($Dist.obs) and a matrix of
the pairwise significance levels based on permutation ($Prob.dist).
3.5
Covariation methods: Two-­‐block partial least squares analysis for shape data (two.b.pls) Function performs two-block partial least squares analysis to assess the degree of association between to
blocks of Procrustes-aligned coordinates (or other variables).
Function:
two.b.pls(A1, A2, warpgrids = TRUE, iter = 999, verbose = FALSE)
Arguments:
A1
A2
iter
warpgrids
verbose
A 2D array (n x [p1 x k]) or 3D array (p1 x k x n) containing GPA-aligned coordinates for Block1
A 2D array (n x [p2 x k]) or 3D array (p2 x k x n) containing GPA-aligned coordinates for Block2
Number of iterations for significance testing
A logical value indicating whether deformation grids for shapes along PC1 should be displayed (only relevant if data
for A1 or A2 [or both] were input as 3D array)
A logical value indicating whether the output is basic or verbose (see Value below)
The function quantifies the degree of association between two blocks of shape data as defined by landmark
coordinates (or one block of shape data and a block of continuous, multivariate data) using partial least
squares (see Rohlf and Corti 2000).
28
An example of 2B-PLS of head shape against diet,
> data(plethShapeFood)
> plethShapeFood$food
[1,]
[2,]
[3,]
[4,]
...
oligoch
-0.333567
-0.333567
-0.333567
-0.333567
gastrop
isopoda
diplop
chilop
acar
araneida
chelon
coleopt
-0.2116501 -0.3740649 -0.2462653 -0.2774757 0.2035222 -0.3740685 -0.3089659 0.71842496
-0.2116501 3.0823626 4.0018112 -0.2774757 -0.5007004 -0.3740685 -0.3089659 0.39181341
-0.2116501 -0.3740649 -0.2462653 3.5516887 -0.5007004 -0.3740685 -0.3089659 -1.39084544
-0.2116501 -0.3740649 -0.2462653 -0.2774757 0.2035222 2.1697963 -0.3089659 0.71842496
> Y.gpa<-gpagen(plethShapeFood$land)
#GPA-alignment
> #2B-PLS between head shape (block 1) and food use data (block 2)
> two.b.pls(Y.gpa$coords,plethShapeFood$food,iter=99)
A plot of PLS scores from Block1 versus Block2 is provided for the first set of PLS axes. Thin-plate spline
deformation grids along these axes are also shown (if data were input as a 3D array and warpgrids=TRUE). If
verbose = TRUE, the function also returns a list containing the PLS scores for the first block of landmarks
($Xscores) and for the second block of landmarks ($Yscores). These can be used to make further plots
(plot()) and shape deformations (plotRefToTarget()).
Protip! Function can be used in many combinations, and is not exclusively for shape data. For example, one
block as shape data for a structure and the other block shape for another structure (but also see
morphol.integr()), or one block as multivariate continuous data (e.g., Precipitation) and other block for
shape data or two blocks of non-shape, multivariate continuous data.
3.6
Morphological Integration methods: Quantify morphological integration between two modules (morphol.integr) Function quantifies the degree of morphological integration between two modules of Procrustes-aligned
coordinates. The function may be used to assess the degree of morphological integration between two
separate structures or between two modules defined within the same landmark configuration.
Function:
morphol.integr(A1, A2, method = c("PLS", "RV"), warpgrids = TRUE,
iter = 999, verbose = FALSE)
Arguments:
A1
A2
A 2D array (n x [p1 x k]) or 3D array (p1 x k x n) containing GPA-aligned coordinates for the first module
A 2D array (n x [p2 x k]) or 3D array (p2 x k x n) containing GPA-aligned coordinates for the second module
29
method
warpgrids
iter
verbose
Method to estimate morphological integration; see below for details
A logical value indicating whether deformation grids for shapes along PLS1 should be displayed (only relevant if
data for A1 or A2 [or both] were input as 3D array)
Number of iterations for significance testing
A logical value indicating whether the output is basic or verbose (method="PLS" only) (see Value below)
Two analytical approaches are currently implemented to assess the degree of morphological integration,
method="PLS" and method="RV".
(default) the function estimates the degree of morphological integration using two-block
partial least squares, or PLS. When used with landmark data, this analysis is referred to as singular warps
analysis (Bookstein et al. 2003). When method="PLS", the scores along the X & Y PLS axes are also
returned, as is a plot of PLS scores from Block1 versus Block2 along the first set of PLS axes. Thin-plate
spline deformation grids along these axes are also shown (if data were input as a 3D array). Note:
deformation grids are displayed for each block of landmarks separately.
method="PLS"
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
> morphol.integr(Y.gpa$coords[1:5,,],Y.gpa$coords[6:12,,],method="PLS",iter=99)
[1] "No specimen names in data matrix 1. Assuming specimens in same order."
[1] "No specimen names in data matrix 2. Assuming specimens in same order."
$PLS.corr
[1] 0.9108678
$pvalue
[1] 0.01
A plot of PLS scores from Block1 versus Block2 is provided for the first set of PLS axes. Thin-plate spline
deformation grids along these axes are also shown (if data were input as a 3D array and warpgrids=TRUE). If
verbose=TRUE, the function also returns a list containing the PLS scores for the first block of landmarks
($Xscores) and for the second block of landmarks ($Yscores). These can be used to make further plots
(plot()) and shape deformations (plotRefToTarget()).
Protip! If the two blocks of landmarks are derived from a single structure (i.e., a single landmark
configuration), one can plot the overall deformation along PLS1 as follows,
30
> res<-morphol.integr(Y.gpa$coords[1:5,,],Y.gpa$coords[6:12,,],
method="PLS",iter=99, verbose=TRUE)
> ref<-mshape(Y.gpa$coords)
#overall reference
> plotRefToTarget(ref,Y.gpa$coords[,,which.min(res$x.scores)],method="TPS") #Min
along PLS1
> plotRefToTarget(ref,Y.gpa$coords[,,which.max(res$x.scores)],method="TPS") #Max
along PLS1
the function estimates the degree of morphological integration using the RV coefficient
(Klingenberg 2009). Significance testing for both approaches is found by permuting the objects in one data
matrix relative to those in the other. A histogram of coefficients obtained via resampling is presented, with
the observed value designated by an arrow in the plot.
method="RV"
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
> morphol.integr(Y.gpa$coords[1:5,,],Y.gpa$coords[6:12,,],method="RV",iter=99)
[1] "No specimen names in data matrix 1. Assuming specimens in same order."
[1] "No specimen names in data matrix 2. Assuming specimens in same order."
$RV
[1] 0.6214977
$pvalue
[1] 0.01
If evaluating an a priori hypothesis of modularity within a structure is of interest, one may use the average
RV coefficient as implemented in the function compare.modular.partitions() 3.7.
31
3.7
Morphological Integration methods: Compare modular signal to alternative landmark subsets (compare.modular.partitions) Function quantifies the degree of morphological integration between two or more modules of Procrustesaligned landmark coordinates defined a priori and compares this to patterns found by randomly assigning
landmarks into subsets. Only modules within the same configuration should be tested.
Function:
compare.modular.partitions(A, landgroups, iter = 999)
Arguments:
A
A 3D array (p x k x n) containing GPA-aligned coordinates for all specimens
landgroups A list of which landmarks belong in which partition (e.g., A,A,A,B,B,B,C,C,C)
iter
Number of iterations for significance testing
The function quantifies the degree of morphological integration between two or more modules of shape data
as defined by landmark coordinates, and compares this to modular signals found by randomly assigning
landmarks to modules. The degree of morphological integration is quantified using the RV coefficient
(Klingenberg 2009). If more than two modules are defined, the average RV coefficient is utilized (see
Klingenberg 2009). The RV coefficient for the observed modular hypothesis is then compared to a
distribution of values obtained by randomly assigning landmarks into subsets, with the restriction that the
number of landmarks in each subset is identical to that observed in each of the original partitions. A
significant modular signal is found when the observed RV coefficient is small relative to this distribution (see
Klingenberg 2009). A histogram of coefficients obtained via resampling is presented, with the observed
value designated by an arrow in the plot.
The landgroups argument can be made by hand, using c(), or there is a graphical assisted function
define.modules().
Function:
define.modules(spec, nmodules)
Arguments:
spec
nmodules
Name of specimen, as an object matrix containing 2D landmark coordinates
Number of modules to be defined
Function takes a matrix of two-dimensional digitized landmark coordinates and allows user assign
landmarks to each module. The output is a list of which landmarks belong in which partition, to be used
by compare.modular.partitions(). The number of modules is chosen by the user (up to five).
define.modules Example where the modularity hypothesis is the cranium versus the mandible,
>
>
>
>
>
>
data(plethodon)
Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
#landmarks on the skull and mandible assigned to partitions
land.gps <- c("A","A","A","A","A","B","B","B","B","B","B","B")
# OR
land.gps <- define.modules(plethodon$land[,,1], nmodules=2)
32
Select landmarks in module 1
Press esc when finished
Select landmarks in module 2
Press esc when finished
[1] "A" "A" "A" "A" "A" "B" "B" "B" "B" "B" "B" "B"
> compare.modular.partitions(Y.gpa$coords,land.gps,iter=99)
> #Result implies that the skull and mandible are not independent modules
3.8
Phylogenetic Comparative methods: Assessing phylogenetic signal in morphometric data (physignal) Function calculates the degree of phylogenetic signal from a set of Procrustes-aligned specimens.
Function:
33
physignal(phy, A, iter = 249, c("Kmult", "SSC"))
Arguments:
phy
A
iter
method
A phylogenetic tree of class 'phylo'
A 2D array (n x [p x k]) or 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens
Number of iterations for significance testing
XXX
The function estimates the degree of phylogenetic signal present in shape data for a given phylogeny. Two
approaches may be used to quantify phylogenetic signal. First, method = "Kmult", a multivariate version of
the K-statistic may be utilized (Kmult: Adams 2014a). This value evaluates the degree of phylogenetic signal
in a dataset relative to what is expected under a Brownian motion model of evolution. For geometric
morphometric data, the approach is a mathematical generalization of the Kappa statistic(Blomberg et al.
2003) appropriate for highly multivariate data (see Adams 2014a).
The second approach, method = "SSC" , estimates the phylogenetic signal as the sum of squared changes
(SSC) in shape along all branches of the phylogeny (Klingenberg and Gidaszewski 2010). Significance
testing is found by permuting the shape data among the tips of the phylogeny. Note that the method can be
slow as ancestral states must be estimated for every iteration.
A plot of the specimens in tangent space with the phylogeny superimposed is included.
The tree must have number of tips equal to number of taxa in the data matrix (e.g., ape package function
drop.tip()). And, tip labels of the tree MUST be exactly the same as the taxa names in the landmark data
matrix (check using match()). To learn more about phylogenetic trees in R, look at these resources:
http://www.r-phylo.org/wiki/ and
http://bodegaphylo.wikispot.org/Phylogenetics_and_Comparative_Methods_in_R
> data(plethspecies)
> Y.gpa<-gpagen(plethspecies$land)
#GPA-alignment
> plethspecies$phy # look at the structure of the tree (class phylo)
Phylogenetic tree with 9 tips and 8 internal nodes.
Tip labels:
P_serratus, P_cinereus, P_shenandoah, P_hoffmani, P_virginia, P_nettingi, ...
Rooted; includes branch lengths.
> plot(plethspecies$phy)
34
Example using method = "Kmult",
> physignal(plethspecies$phy,Y.gpa$coords,iter=99, method = "Kmult")
$phy.signal
[,1]
[1,] 0.957254
$pvalue
[,1]
[1,] 0.02
Example using method = “SSC” ,
> physignal(plethspecies$phy,Y.gpa$coords,iter=99, method = "SSC")
$phy.signal
[1] 0.002235759
$pvalue
[1] 0.01
This function also calls plotGMPhyloMorphoSpace()4.3 and plots the phylomorphospace.
This function can be used with univariate data (i.e., centroid size) if imported as matrix with row names
giving the taxa names.
># make matrix Csize with names
> Csize <- matrix(Y.gpa$Csize, dimnames=list(names(Y.gpa$Csize)))
> physignal(plethspecies$phy,Csize,iter=99, method = "Kmult")
$phy.signal
[,1]
[1,] 0.7097642
$pvalue
[,1]
[1,] 0.49
> physignal(plethspecies$phy,Csize,iter=99, method = "SSC")
$phy.signal
[1] 1.762033e-07
$pvalue
[1] 0.3
35
3.9
Phylogenetic Comparative methods: Quantify phylogenetic morphological integration between two sets of variables (phylo.pls) Function quantifies the degree of phylogenetic morphological covariation between two sets of Procrustesaligned coordinates using partial least squares.
Function:
phylo.pls(A1, A2, phy, warpgrids = TRUE, iter = 999, verbose = FALSE)
Arguments:
A1
A2
phy
warpgrids
iter
verbose
A 2D array (n x [p1 x k]) or 3D array (p1 x k x n) containing landmark coordinates for the first block
A 2D array (n x [p2 x k]) or 3D array (p2 x k x n) containing landmark coordinates for the second block
Object of class ‘phylo’ with tip labels corresponding to the row names of the blocks of data
A logical value indicating whether deformation grids for shapes along PC1 should be displayed (only relevant if data
for A1 or A2 [or both] were input as 3D array)
Number of iterations for significance testing
A logical value indicating whether the output is basic or verbose (see Value below)
The function estimates the degree of morphological covariation between two sets of variables while
accounting for phylogeny using partial least squares (Adams and Felice 2014). The observed value is
statistically assessed using permutation, where data for one block are permuted across the tips of the
phylogeny, an estimate of the covariation between sets of variables, and compared to the observed value. A
plot of PLS scores from Block1 versus Block2 is provided for the first set of PLS axes. Thin-plate spline
deformation grids along these axes are also shown (if data were input as a 3D array).
> data(plethspecies)
> Y.gpa<-gpagen(plethspecies$land)
#GPA-alignment
> phylo.pls(Y.gpa$coords[1:5,,],Y.gpa$coords[6:11,,],plethspecies$phy,iter=5)
$`PLS Correlation`
[1] 0.9338463
$pvalue
[,1]
[1,] 0.02777778
A plot of PLS scores from Block1 versus Block2 is provided for the first set of PLS axes. Thin-plate spline
deformation grids along these axes are also shown (if data were input as a 3D array and warpgrids=TRUE). If
verbose=TRUE, the function also returns a list containing the PLS scores for the first block of landmarks
($Xscores) and for the second block of landmarks ($Yscores). These can be used to make further plots
(plot()) and shape deformations (plotRefToTarget()).
36
3.10
Phylogenetic Comparative methods: Comparing rates of shape evolution on phylogenies (compare.evol.rates) Function calculates rates of shape evolution for two or more groups of species on a phylogeny from a set of
Procrustes-aligned specimens.
Function:
compare.evol.rates(phy, A, gp, iter = 999)
Arguments:
phy
A
gp
iter
A phylogenetic tree of class 'phylo'
A matrix (n x [p x k]) or 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens
A factor array designating group membership
Number of iterations for significance testing
The function compares rates of morphological evolution for two or more groups of species on a phylogeny,
under a Brownian motion model of evolution. The approach is based on the distances between species in
morphospace after phylogenetic transformation (Adams 2014b). From the data the rate of shape evolution for
each group is calculated, and a ratio of rates is obtained. If three or more groups of species are used, the ratio
of the maximum to minimum rate is used as a test statistic (see Adams 2014b). Significance testing is
accomplished by phylogenetic simulation in which tips data are obtained under Brownian motion using a
single evolutionary rate for all species on the phylogeny. If three or more groups of species are used, pairwise
P-values are also returned. A histogram of evolutionary rate ratios obtained via phylogenetic simulation is
presented, with the observed value designated by an arrow in the plot. The function can be used to obtain a
rate for the whole dataset of species by using a dummy group factor assigning all species to one group.
Example, comparing endangered versus not endangered species of Plethodon,
> data(plethspecies)
> Y.gpa<-gpagen(plethspecies$land)
#GPA-alignment
> gp.end<-factor(c(0,0,1,0,0,1,1,0,0)) #endangered species vs. rest
> names(gp.end)<-plethspecies$phy$tip
> compare.evol.rates(plethspecies$phy,Y.gpa$coords,gp=gp.end,iter=49)
$sigma.d
[1] 2.297943e-06
$sigmad.all
0
1
1.796579e-06 3.300672e-06
$sigmad.ratio
[1] 1.837199
$pvalue
[1] 0.02
This function can be used with univariate data (i.e., centroid size) if imported as matrix with row names
giving the taxa names,
> Csize <- matrix(Y.gpa$Csize, dimnames=list(names(Y.gpa$Csize))) # make matrix
Csize with names
37
> compare.evol.rates(plethspecies$phy,Csize,gp=gp.end,iter=49)
$sigma.d
[1] 3.013353e-09
$sigmad.all
0
1
1.776216e-09 5.487627e-09
$sigmad.ratio
[1] 3.089505
$pvalue
[1] 0.38
3.11 Calculate morphological disparity for one or more groups (morphol.disparity) Function estimates morphological disparity and performs pairwise comparisons to identify differences
between groups.
Function:
morphol.disparity(A, groups, iter = 999)
Arguments:
A
groups
iter
A 2D array (n x [p x k]) or 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens
A factor defining groups
Number of iterations for permutation test
The function takes as input GPA-aligned shape data and a grouping factor, and estimates disparity as the
Procrustes variance for each group, which is the sum of the diagonal elements of the group covariance matrix
(Zelditch et al. 2012). The group Procrustes variances are used as test values, and these are then statistically
evaluated through permutation, where the rows of the shape matrix are randomized relative to the grouping
variable. The function can be used to obtain disparity for the whole dataset by using a dummy group factor
assigning all specimens to one group, in which case only Procrustes variance is returned.
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
> morphol.disparity(Y.gpa$coords, groups=plethodon$site, iter = 99)
$Disp.obs
ProcVar
Allo 0.001981131
Symp 0.004647113
$Prob.Disp
Allo Symp
Allo 1.00 0.01
Symp 0.01 1.00
Use plotTangentSpace() to view the morphospace and color by group.
38
3.12 Quantify and compare shape change trajectories (trajectory.analysis) Function estimates attributes of shape change trajectories or motion trajectories for a set of Procrustesaligned specimens and compares them statistically.
Function:
trajectory.analysis(f1, data = NULL, estimate.traj = TRUE,
traj.pts = NULL, iter = 99)
Arguments:
f1
data
estimate.traj
iter
traj.pts
A formula for the linear model (e.g., y~x1+x2), where y is a two-dimensional array of shape data
An optional value specifying a data frame containing all data (not required)
A logical value indicating whether trajectories are estimated from original data; described below
Number of iterations for significance testing
An optional value specifying the number of points in each trajectory (if estimate.traj=FALSE)
The function quantifies phenotypic shape change trajectories from a set of specimens, and assesses variation
in these parameters via permutation. A shape change trajectory is defined by a sequence of shapes in tangent
space. These trajectories can be quantified various attributes (their size, orientation, and shape), and
comparisons of these attribute enables the statistical comparison of shape change trajectories (see Adams and
Collyer 2007; Collyer and Adams 2007; Adams and Collyer 2009; Collyer and Adams 2013).
Data input is specified by a formula (e.g., Y~X), where 'Y' specifies the response variables (trajectory data),
and 'X' contains one or more independent variables (discrete or continuous). The response matrix 'Y' must be
in the form of a two-dimensional data matrix of dimension (n x [p x k]), rather than a 3D array. The function
two.d.array() can be used to obtain a two-dimensional data matrix from a 3D array of landmark
coordinates. It is assumed that the order of the specimens 'Y' matches the order of specimens in 'X'.
There are two primary modes of analysis through this function. If "estimate.traj=TRUE" the function
estimates shape trajectories using the least-squares means for groups, based on a two-factor model (e.g.,
Y~A+B+A:B). Under this implementation, the last factor in 'X' must be the interaction term, and the
preceding two factors must be the effects of interest. Covariates may be included in 'X', and must precede the
factors of interest (e.g., Y~cov+A*B). In this implementation, 'Y' contains a matrix of landmark coordinates.
It is assumed that the landmarks have previously been aligned using Generalized Procrustes Analysis (GPA).
If "estimate.traj=FALSE" the trajectories are assembled directly from the set of shapes provided in 'Y'.
With this implementation, the user must specify the number of shapes that comprise each trajectory. This
approach is useful when the set of shapes forming each trajectory have been quantified directly (e.g., when
motion paths are compared: see Adams and Cerney 2007). With this implementation, variation in trajectory
size, shape, and orientation are evaluated for each term in 'X' (see Adams and Cerney 2007).
Once the function has performed the analysis, it generates a plot of the trajectories as visualized in the space
of principal components (PC1 vs. PC2). The first point in each trajectory is displayed as white, the last point
is black, and any middle points on the trajectories are in gray. The colors of trajectories follow the order in
which they are found in the dataset, using R's standard color palette: black, red, green, blue, cyan, magenta,
yellow, and gray.
Example to estimate trajectories from LS means in 2-factor model
> data(plethodon)
> Y.gpa<-two.d.array(gpagen(plethodon$land)$coords)
> trajectory.analysis(Y.gpa~plethodon$species*plethodon$site,iter=15)
[1] "No specimen names in response matrix. Assuming specimens in same order."
$ProcDist.lm
df
SS.obs
MS
F P.val
Rsq
plethodon$species
1 0.02925784 0.029257838 14.54367 0.0625 0.1485624
39
plethodon$site
1 0.06437484 0.064374844 31.99985 0.0625 0.3268759
plethodon$species:plethodon$site 1 0.03088502 0.030885020 15.35252 0.0625 0.1568247
Total
39 0.19693973 0.005049737
NA
NA
NA
$traj.size
1
2
1 0.000000000 0.003831272
2 0.003831272 0.000000000
$p.size
1
2
1 1.0 0.5
2 0.5 1.0
$traj.orient
[,1]
[,2]
[1,] 0.00000 69.40048
[2,] 69.40048 0.00000
$p.orient
[,1]
[,2]
[1,] 1.0000 0.0625
[2,] 0.0625 1.0000
Compare motion trajectories
> data(motionpaths)
> motionpaths #Motion paths represented by 5 time points per motion
$trajectories
[,1]
[,2]
[1,] 1.5547554 10.882688
[2,] 3.1034540 11.794735
...
[,3]
[,4]
2.292724 13.795449
2.445819 13.830566
[,5]
[,6]
3.189067 16.425534
3.197470 18.280553
[,7]
[,8]
5.576099 17.579038
6.432909
$groups
[1] 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4
Levels: 1 2 3 4
> trajectory.analysis(motionpaths$trajectories~motionpaths$groups,
estimate.traj=FALSE, traj.pts=5,iter=15)
[1] "No specimen names in response matrix. Assuming specimens in same order."
$MANOVA.location.covariation
df
SS.obs
MS
F P.val
Rsq
motionpaths$groups 3 6520.905 2173.6349 849.846 0.0625 0.9860764
Total
39 6612.981 169.5636
NA
NA
NA
$ANOVA.Size
Df SumsOfSqs MeanSqs F.Model
R2 Pr(>F)
motionpaths$groups 3
103.345 34.448 51.392 0.8107 0.0625 .
Residuals
36
24.131
0.670
0.1893
Total
39
127.475
1.0000
--Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
40
$ANOVA.Dir
Df SumsOfSqs MeanSqs F.Model
R2 Pr(>F)
motionpaths$groups 3
20490.4 6830.1 90.147 0.88252 0.0625 .
Residuals
36
2727.6
75.8
0.11748
Total
39
23218.0
1.00000
--Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
41
Section 4: Visualization 4.1 Principal Components Analysis (plotTangentSpace) Function plots a set of Procrustes-aligned specimens in tangent space along their principal axes, i.e., function
performs a Principal Component Analysis (PCA).
Function:
plotTangentSpace(A, axis1 = 1, axis2 = 2, warpgrids = TRUE, mesh = NULL,
label = FALSE, groups = NULL, verbose = FALSE)
Arguments:
A
warpgrids
mesh
axis1
axis2
label
groups
verbose
A 3D array (p x k x n) containing landmark coordinates for a set of aligned specimens
A logical value indicating whether deformation grids for shapes along X-axis should be displayed
A mesh3d object to be warped to represent shape deformation along X-axis (when warpgrids=TRUE) as
described in warpRefMesh().
A value indicating which PC axis should be displayed as the X-axis (default = PC1)
A value indicating which PC axis should be displayed as the X-axis (default = PC2)
A logical value indicating whether labels for each specimen should be displayed
An optional factor vector specifying group identity for each specimen
A logical value indicating whether the output is basic or verbose
The function performs a principal components analysis of shape variation and plots two dimensions of
tangent space for a set of Procrustes-aligned specimens (default is PC1 vs. PC2). The percent variation along
each PC-axis is returned. Additionally (and optionally, warpgrids=T), deformation grids can be requested,
which display the shape of specimens at the ends of the range of variability along PC1. If groups are
provided, specimens from each group are plotted using distinct colors based on the order in which the groups
are found in the dataset, and using R's standard color palette: black, red, green, blue, cyan, magenta, yellow,
and gray.
>
>
>
>
data(plethodon)
Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
ref<-mshape(Y.gpa$coords)
plotTangentSpace(Y.gpa$coords, groups = paste(plethodon$species,
plethodon$site))
Importance of components:
PC1
PC2
PC3
PC4
PC5
PC6
PC7
PC8
Standard deviation
0.04307 0.03958 0.02035 0.01509 0.01314 0.01293 0.01245 0.01119
Proportion of Variance 0.36743 0.31023 0.08201 0.04512 0.03418 0.03312 0.03068 0.02478
Cumulative Proportion 0.36743 0.67767 0.75967 0.80479 0.83897 0.87209 0.90277 0.92755
PC9
PC10
PC11
PC12
PC13
PC14
PC15
Standard deviation
0.009202 0.008289 0.007369 0.006651 0.005116 0.004389 0.004173
Proportion of Variance 0.016770 0.013600 0.010750 0.008760 0.005180 0.003810 0.003450
Cumulative Proportion 0.944320 0.957920 0.968670 0.977430 0.982620 0.986430 0.989880
PC16
PC17
PC18
PC19
PC20
PC21
PC22
Standard deviation
0.004141 0.003944 0.003042 0.00231 0.00195 1.401e-15 2.802e-16
Proportion of Variance 0.003400 0.003080 0.001830 0.00106 0.00075 0.000e+00 0.000e+00
Cumulative Proportion 0.993280 0.996360 0.998190 0.99925 1.00000 1.000e+00 1.000e+00
PC23
PC24
Standard deviation
3.426e-17 1.041e-17
Proportion of Variance 0.000e+00 0.000e+00
Cumulative Proportion 1.000e+00 1.000e+00
42
When verbose=TRUE, the function also returns a list contacting the PC scores ($pc.scores) and the
minimum and maximum shapes ($pc.shapes) on the two PCs plotted, which can be used in
plotRefToTarget().
$pc.scores
PC1
PC2
PC3
PC4
PC5
PC6
[1,] -0.0369931316 0.051182469 -0.0016971188 -3.128812e-03 -0.0109363280 -0.0264503045
[2,] -0.0007493756 0.059420824 0.0001371746 -2.768676e-03 -0.0081174155 0.0139162571
..
$pc.shapes
$pc.shapes$PC1min
[,1]
[,2]
[1,] 0.16960517 -0.02807690
[2,] 0.21810341 -0.10340535
...
$pc.shapes$PC1max
[,1]
[,2]
[1,] 0.14235007 -0.020486784
[2,] 0.18012044 -0.086122720
...
$pc.shapes$PC2min
[,1]
[,2]
[1,] 0.13128088 -0.03616776
[2,] 0.18767216 -0.10979356
...
$pc.shapes$PC2max
[,1]
[,2]
[1,] 0.18546661 -0.003640617
[2,] 0.20487416 -0.066270202
...
43
mshape Function:
mshape(A)
Arguments:
A An array (p x k x n) containing GPA-aligned coordinates for a set of specimens
The function estimates the average landmark coordinates for a set of aligned specimens. This function is
described in Claude (2008).
> mshape(Y.gpa$coords)
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
[9,]
[10,]
[11,]
[12,]
[,1]
0.15263286
0.19445064
-0.03361223
-0.28069721
-0.30998751
-0.32557841
-0.31804390
-0.18947114
0.02036830
0.18852641
0.35134313
0.55006906
[,2]
-0.023350382
-0.092643105
-0.007346224
-0.092850148
-0.061672266
-0.036112257
0.036125317
0.098010904
0.099112884
0.077278070
0.065866300
-0.062419094
4.2 Plot allometric patterns in landmark data (plotAllometry) Function performs a regression and plots allometry curves for a set of specimens.
Function:
plotAllometry(A, sz, groups = NULL, method = c("CAC", "RegScore",
"PredLine"), warpgrids = TRUE, iter = 99, label = FALSE, mesh = NULL,
verbose = FALSE)
Arguments:
A
sz
groups
method
warpgrids
mesh
iter
label
verbose
A 3D array (p x k x n) containing landmark coordinates for a set of specimens
A vector of centroid size measures for all specimens
An optional vector containing group labels for each specimen if available
Method for estimating allometric shape components; see below for details
A logical value indicating whether deformation grids for small and large shapes should be displayed
A mesh3d object to be warped to represent shape deformation of the directional and fluctuating components of
asymmetry if warpgrids= TRUE (as described in warpRefMesh()).
Number of iterations for significance testing
A logical value indicating whether labels for each specimen should be displayed
A logical value indicating whether the output is basic or verbose
The function performs a regression of shape on size, and generates a plot that describes the multivariate
relationship between size and shape derived from landmark data (i.e., allometry). The abscissa of the plot is
log(centroid size) while the ordinate represents shape. Three complementary approaches can be implemented
to visualize allometry, "method=CAC", "method=RegScore", "method=PredLine". For all methods, both
centroid size and allometry scores are returned. Optionally, deformation grids can be requested, which
display the shape of the smallest and largest specimens relative to the average specimen (using
'warpgrids=T' or 'warpgrids=F'). Finally, if groups are provided, the above approaches are implemented
while accounting for within-group patterns of covariation (see references for explanation). In this case, the
44
regression is of the form: shape~size+groups (Note: to examine the interaction term use procD.lm()).
Specimens from each group are plotted using distinct colors based on the order in which the groups are found
in the dataset, and using R's standard color palette: black, red, green, blue, cyan, magenta, yellow, and gray.
If "method=CAC" (the default) the function calculates the common allometric component of the shape data,
which is an estimate of the average allometric trend within groups (Mitteroecker et al. 2004). The function
also calculates the residual shape component (RSC) for the data.
> data(ratland)
> Y.gpa<-gpagen(ratland)
#GPA-alignment
> plotAllometry(Y.gpa$coords,Y.gpa$Csize,method="CAC", iter=5)
[1] "No specimen names in data matrix. Assuming specimens in same order."
[1] "No specimen names in size vector. Assuming specimens in same order."
[1] "No specimen names in response matrix. Assuming specimens in same order."
$ProcDist.lm
df
SS.obs
MS
F
P.val
Rsq
csz
1 0.6280794 0.628079386 474.7793 0.1666667 0.7455947
Total 163 0.8423871 0.005168019
NA
NA
NA
If "method=RegScore" the function calculates shape scores from the regression of shape on size, and plots
these versus size (Drake and Klingenberg 2008). For a single group, these shape scores are mathematically
identical to the CAC (Adams et al. 2013).
> plotAllometry(Y.gpa$coords,Y.gpa$Csize,method="RegScore", iter=5)
[1] "No specimen names in data matrix. Assuming specimens in same order."
[1] "No specimen names in size vector. Assuming specimens in same order."
[1] "No specimen names in response matrix. Assuming specimens in same order."
$ProcDist.lm
df
SS.obs
MS
F
P.val
Rsq
csz
1 0.6280794 0.628079386 474.7793 0.1666667 0.7455947
Total 163 0.8423871 0.005168019
NA
NA
NA
45
If "method=PredLine" the function calculates predicted values from a regression of shape on size, and plots
the first principal component of the predicted values versus size as a stylized graphic of the allometric trend
(Adams and Nistri 2010)
> plotAllometry(Y.gpa$coords,Y.gpa$Csize,method="PredLine", iter=5)
[1] "No specimen names in data matrix. Assuming specimens in same order."
[1] "No specimen names in size vector. Assuming specimens in same order."
[1] "No specimen names in response matrix. Assuming specimens in same order."
$ProcDist.lm
df
SS.obs
MS
F
P.val
Rsq
csz
1 0.6280794 0.628079386 474.7793 0.1666667 0.7455947
Total 163 0.8423871 0.005168019
NA
NA
NA
When verbose=TRUE, the function returns a list containing the ANOVA table ($ProcDist.lm) a matrix of
allometry shape scores ($allom.score), a matrix of log centroid size ($LogCsize), a matrix of the predicted
shapes for the regression ($pred.shape) – the latter of which can be used with plotRefToTarget(). If
"method=CAC" the list also has the residual shape component (RSC) ($resid.shape).
Protip! This function is designed to show shape variation with log(centroid size) for allometric study
questions. It is however possible to use this function to view shape changes associated with any other
continuous variable (e.g., body length, or precipitation). The function automatically takes the natural log of
the x-variable. Therefore when using this function for variables that should not be logged, simply input them
46
as the exponent (exp(x)) and ignore the x-axis graph labels. The allometric shape scores are returned when
verbose=T, and these can be used to plot the graph how the user wishes.
4.3 Plot phylogenetic tree and specimens in tangent space (plotGMPhyloMorphoSpace) Function plots a phylogenetic tree and a set of Procrustes-aligned specimens in tangent space, i.e., a
phylomorphospace. This function is also called within physignal().
Function:
plotGMPhyloMorphoSpace(phy, A, labels = TRUE, ancStates = TRUE)
Arguments:
phy
A
labels
ancStates
A phylogenetic tree of class phylo - see read.tree in library ape
A matrix (n x [p x k]) or 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens
A logical value indicating whether taxa labels should be included
A logical value indicating whether ancestral state values should be returned
The function creates a plot of the first two dimensions of tangent space for a set of Procrustes-aligned
specimens. The phylogenetic tree for these specimens is superimposed in this plot revealing how shape
evolves (e.g., Rohlf 2002; Klingenberg and Gidaszewski 2010). The plot also displays the ancestral states for
each node of the phylogenetic tree, whose values can optionally be returned (ancStates = TRUE).
> data(plethspecies)
> Y.gpa<-gpagen(plethspecies$land)
#GPA-alignment
> plotGMPhyloMorphoSpace(plethspecies$phy,Y.gpa$coords)
[1,]
[2,]
[3,]
[4,]
[5,]
...
[,1]
[,2]
[,3]
0.2113723 0.0033556111 0.2511653
0.2141588 0.0012264326 0.2549009
0.2159035 0.0014208297 0.2587431
0.2152061 -0.0009607547 0.2547973
0.2155224 -0.0005148555 0.2553910
[,4]
-0.05389905
-0.05405432
-0.05061208
-0.05761135
-0.05443169
[,5]
-0.02184034
-0.02194802
-0.01859938
-0.02540284
-0.02805191
[,6]
-0.01610523
-0.01717149
-0.01730399
-0.01809403
-0.01748467
4.4 Create a mesh3d object warped to the mean shape (warpRefMesh) A function to take a ply file and use thin-plate spline method to warp the file into the estimated mean shape
for a set of aligned specimens. This mesh is used in functions where mesh= option is available.
Function:
warpRefMesh(file, mesh.coord, ref, color = NULL, centered = FALSE)
Arguments:
47
file
mesh.coord
ref
color
centered
An ASCII ply file
A p x k matrix of 3D coordinates digitized on the ply file.
A p x k matrix of 3D coordinates made by mshape()
Color to set the ply file $material. If the ply already has color, use NULL. For ply files without color, color=NULL
will be plotted as grey.
Logical If the data in mesh.coords were collected from a centered mesh (see below).
Function takes a ply file and its landmark coordinates uses the thin-plate spline method (Bookstein 1989) to
warp the mesh into the shape defined by a second set of landmark coordinates, usually those of the mean
shape for a set of aligned specimens. It is highly recommended that the mean shape is used as the reference
for warping (see Rohlf 1998).
The workflow is as follows:
1. Calculate the mean shape using mshape()
2. Choose an actual specimen to use for the warping. The specimen used as the template for this
warping is recommended as one most similar in shape to the average of the sample, but can be any
reasonable specimen – do this by eye, or use findMeanSpec()
3. Warp this specimen into the mean shape using warpRefMesh()
4. Use this average mesh where it asks for a mesh= in the analysis functions and visualization functions
For landmark coordinates digitized with geomorph digitizing functions centered=TRUE by default. This
refers to the specimen being centered prior to landmark acquisition in the RGL window. For landmark data
collected outside of geomorph, centered=FALSE will usually be the case. The returned mesh3d object is for
use in geomorph functions where shape deformations are plotted and mesh= option is available
(plotTangentSpace(), plotAllometry(), bilat.symmetry(), and plotRefToTarget()).
For example,
> #1
> average <- mshape(Y.gpa$coords) # calculate the mean shape
> #2
> findMeanSpec(Y.gpa$coords) # Identify specimen closest to mean shape (input is a
3D array of GPA-aligned coordinates)
specimen7
# returns the name of the specimen
25
# returns the specimen number (where it appears in the 3D array)
> # 3
filelist = c("specimen7.nts","specimen7.nts") # makes false list so a single nts
file can be read in using readmulti.nts
> specimen7 <- readmulti.nts(filelist)
remove(filelist) # read in the nts file, then remove the unnecessary objects
># 4
> averagemesh <- warpRefMesh("specimen7.ply", specimen7[,,1], average, color=NULL)
The function returns a mesh3d object of the mesh warped to the shape of the average. It also plots the
imported mesh (here “specimen7”) and the new warped mesh:
48
Now the object “averagemesh” from our example above can be used in geomorph functions where shape
deformations are plotted (plotTangentSpace(), plotAllometry(), bilat.symmetry(), and
plotRefToTarget()).
The “averagemesh” can also be saved to the working directory using the rgl function writePLY().
findMeanSpec Function:
findMeanSpec(A)
A function to identify which specimen lies closest to the estimated mean shape for a set of aligned
specimens. A is a 3D array (p x k x n) containing landmark coordinates for a set of aligned specimens.
This function is used to facilitate finding a specimen to use with warpRefMesh().
> findMeanSpec(Y.gpa$coords)
specimen7
# returns the name of the specimen
25
# returns the specimen number (where it appears in the 3D array)
4.5 Plot landmark coordinates for all specimens (plotAllSpecimens) Function plots landmark coordinates for a set of specimens.
Function:
plotAllSpecimens(A, mean = TRUE, links = NULL, pointscale = 1, meansize = 2)
Arguments:
A
mean
links
pointscale
meansize
A 3D array (p x k x n) containing GPA-aligned coordinates for a set of specimens
A logical value indicating whether the mean shape should be included in the plot
An optional matrix defining for links between landmarks
An optional value defining the size of the points for all specimens
An optional value defining the size of the points representing the average specimen
49
The function creates a plot of the landmark coordinates for all specimens. This is useful for examining
patterns of shape variation after GPA. If "mean=TRUE", the mean shape will be calculated and added to the
plot. Additionally, if a matrix of links is provided, the landmarks of the mean shape will be connected by
lines. The link matrix is an m x 2 matrix, where m is the desired number of links. Each row of the link matrix
designates the two landmarks to be connected by that link. The function will plot either two- or threedimensional data.
Example for 2D data
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
> plethodon$links
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
[9,]
[10,]
[11,]
[12,]
[13,]
[14,]
[,1] [,2]
4
5
3
5
2
4
1
2
1
3
6
7
7
8
8
9
9
10
10
11
11
12
12
1
1
9
1
10
> plotAllSpecimens(Y.gpa$coords,links=plethodon$links)
Example for 3D data
> data(scallops)
> gpagen(A=scallops$coorddata, curves=scallops$curvslide,
surfaces=scallops$surfslide)
Protip! Typing the name of the function in the console brings up the function code. The user can look at this
code for ideas of how to customize their own graphs.
4.6 Plot shape differences between a reference and target specimen (plotRefToTarget) Function plots shape differences between a reference and target specimen.
Function:
plotRefToTarget(M1, M2, mesh = NULL, method = c("TPS", "vector", "points",
50
"surface"), mag = 1, links = NULL, ...)
Arguments:
M1
M2
mesh
method
mag
links
...
Matrix of landmark coordinates for the first (reference) specimen
Matrix of landmark coordinates for the second (target) specimen
A mesh3d object for use with method="surface"
Method used to visualize shape difference; see below for details
The desired magnification to be used when visualizing the shape difference (e.g., mag=2)
An optional matrix defining for links between landmarks
Additional parameters to be passed to plot, plot3d or shade3d.
The function generates a plot of the shape differences of a target specimen relative to a reference specimen.
The option mag allows the user to indicates the degree of magnification to be used when displaying the shape
difference. The function will plot either two- or three-dimensional data. This function combines numerous
plotting functions found in Claude (2008). Four methods for plots are available, TPS, vector, points,
surface.
The function can be used to show shape deformation using shape coordinates from most analyses. Below are
examples to visualize shape deformations from the mean shape to another specimen in the dataset.
A 2D data example
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
> ref<-mshape(Y.gpa$coords)
#GPA-alignment
a thin-plate spline deformation grid is generated. For 3D data, this method will generate thin-plate spline
deformations in the x-y and x-z planes.
TPS
> plotRefToTarget(ref,Y.gpa$coords[,,39], method="TPS")
> plotRefToTarget(ref,Y.gpa$coords[,,39],mag=3, method="TPS")
by 3X
#magnify difference
a plot showing the vector displacements between corresponding landmarks in the reference and
target specimen is shown.
vector
> plotRefToTarget(ref,Y.gpa$coords[,,39],method="vector", mag=3)
51
a plot is displayed with the landmarks in the target (black) overlaying those of the reference (gray).
Additionally, if a matrix of links is provided, the landmarks of the mean shape will be connected by lines.
The link matrix is an M x 2 matrix, where M is the desired number of links. Each row of the link matrix
designates the two landmarks to be connected by that link.
points
> plotRefToTarget(ref,Y.gpa$coords[,,39],method="points", mag=3)
A 3D data example using the three methods shown above
> data(scallops)
> Y.gpa<-gpagen(A=scallops$coorddata, curves=scallops$curvslide,
surfaces=scallops$surfslide)
> ref<-mshape(Y.gpa$coords)
> plotRefToTarget(ref,Y.gpa$coords[,,1],method="TPS", mag=3)
52
> plotRefToTarget(ref,Y.gpa$coords[,,1],method="vector", mag=3)
> plotRefToTarget(ref,Y.gpa$coords[,,1],method="points", mag=3)
surface a mesh3d surface is warped using thin-plate spline (for 3D data only). Requires mesh3d object in
option mesh=, preferably made using warpRefMesh() 4.4, which provides a mesh3d object that is the shape
of the sample mean. It is recommended that the mean shape is used as the reference for warping (see Rohlf
1998).
A 3D data example using the average mesh made with warpRefMesh()
(Note: this example is not included in the geomorph package, but show here for illustrative purposes).
> # averagemesh is a mesh made with warpRefMesh, that is the shape of the mean of a
set of specimens
> ref <- mshape(Y.gpa$coords) # calculate the mean shape from a set of GPA-aligned
specimens
> plotRefToTarget(M1=ref, M2=Y.gpa$coords[,,1], mesh=averagemesh, method="surface")
Function plots the warped “target” shape (shown here against the mean for illustrative purposes only).
This function can be used to show deformations between any two sets of coordinates. Coordinate data are
provided by several functions, a few examples given below.
53
Example 1, for plotTangentSpace() verbose=TRUE returns a list containing shape coordinates
($pc.shapes) for the minimum and maximum shape on the two PCs plotted (default is PC1 and PC2) as four
matrices. To use, enter the coordinate matrix into position M2, e.g.
> data(plethodon)
> Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
> ref<-mshape(Y.gpa$coords)
> res <- plotTangentSpace(Y.gpa$coords, groups = paste(plethodon$species,
plethodon$site), verbose=TRUE)
> plotRefToTarget(M1=ref, M2=res$pc.shapes$PC1min, method="TPS") #shape change along
PC1 in the negative direction
> plotRefToTarget(M1=ref, M2=res$pc.shapes$PC1max, method="TPS") #shape change along
PC1 in the positive direction
Example 2, for plotAllometry() verbose=TRUE returns a list containing the predicted shape scores
($pred.shape). To use, plot the minimum or maximum of this array in M2, e.g.,
> data(ratland)
> Y.gpa<-gpagen(ratland)
#GPA-alignment
> ref<-mshape(Y.gpa$coords)
> res <- plotAllometry(Y.gpa$coords,Y.gpa$Csize,method="RegScore", verbose=TRUE)
> plotRefToTarget(ref, A$pred.shape[,,which.max(Y.gpa$Csize)],method="TPS") #
Predicted shape at max centrod size
> plotRefToTarget(ref, A$pred.shape[,,which.min(Y.gpa$Csize)],method="TPS") #
Predicted shape at min centrod size
Example 3, for two.b.pls(), morphol.integr( method=”PLS”) and phylo.pls() verbose=TRUE
returns the PLS scores for block 1 and block 2 ($Xscores, $Yscores). M2 plots the actual specimen that is
at the maximum or minimum of block 1. For example, taking the minimum and maximum along the Xblock
1 PLS 1, e.g.,
> res<morphol.integr(Y.gpa$coords[1:5,,], Y.gpa$coords[6:12,,],
method="PLS",iter=99,verbose=TRUE)
> ref<-mshape(Y.gpa$coords)
> plotRefToTarget(ref,Y.gpa$coords[,,which.min(res$x.scores)],method="TPS") #Min
along PLS1
> plotRefToTarget(ref,Y.gpa$coords[,,which.max(res$x.scores)],method="TPS") #Max
along PLS1
Example 4, for viewing differences between groups (shown to be significantly different with procD.lm()),
first average the data by groups, calculate group means and plot those means.
>
>
>
>
>
data(plethodon)
Y.gpa<-gpagen(plethodon$land)
#GPA-alignment
y <- two.d.array(Y.gpa$coords)
means <- aggregate(y ~ plethodon$site, FUN=mean)
means
plethodon$site
V1
V2
V3
V4
V5
V6
V7
V8
1
Allo 0.1397395 -0.02535584 0.1832314 -0.09279558 -0.02462463 -0.006231783 -0.2797076 0.09041826
2
Symp 0.1655262 -0.02134492 0.2056699 -0.09249063 -0.04259982 -0.008460664 -0.2816868 0.09528204
54
...
> Allo.mn <- matrix(as.numeric(means[1,-1]), ncol=2, byrow=T) #makes as matrix
> ref<-mshape(Y.gpa$coords)
> plotRefToTarget(ref, Allo.mn, method="TPS") # Means shape of group
4.7 Plot 3D specimen, fixed landmarks and surface semilandmarks (plotspec) A function to plot three-dimensional (3D) specimen along with its landmarks.
Function:
plotspec(spec, digitspec, fixed = NULL, ptsize = 1, centered = FALSE, ...)
Arguments:
spec
digitspec
fixed
ptsize
centered
...
An object of class shape3d/mesh3d, or matrix of 3D vertex coordinates.
Name of data matrix containing 3D fixed and/or surface sliding coordinates.
Numeric The number of fixed template landmarks (listed first in digitspec)
Numeric Size to plot the mesh points (vertices), e.g., 0.1 for dense meshes, 3 for sparse meshes
Logical Whether the data matrix is in the surface mesh coordinate system (centered=FALSE) or if the data were
collected after the mesh was centered (centered=TRUE)- see details.
additional parameters which will be passed to plot3d.
Function to plot 3D specimens along with their digitized "fixed" landmarks and semilandmarks "surface
sliders" and "curve sliders". If specimen is a 3D surface (class shape3d/mesh3d) mesh is plotted. For
visualization purposes, 3D coordinate data collected using digit.fixed() or digitsurface() and
buildtemplate() prior to build 1.1-6 were centered by default. Therefore use this function with
centered=TRUE. Data collected outside geomorph should be read using centered=FALSE. The function
assumes the fixed landmarks are listed at the beginning of the coordinate matrix (digitspec).
>
>
>
>
data(scallopPLY)
ply <- scallopPLY$ply
digitdat <- scallopPLY$coords
plotspec(spec=ply,digitspec=digitdat,fixed=16, centered =TRUE)
The fixed landmarks and curve semilandmarks are in
red, and the surface semilandmarks are in green.
55
Section 5: Data Collection (Digitizing) 5.1 2D data collection (digitize2d) An interactive function to digitize two-dimensional landmarks from .jpg files.
Function:
digitize2d(file, nlandmarks, scale, verbose = TRUE)
Arguments:
file
nlandmarks
scale
verbose
Name of jpeg file to be digitized. File names can be written in manually, including paths, or obtained using
directory/file manipulation functions e.g., list.files
Number of landmarks to be digitized.
Length of scale placed in image.
User decides whether to digitize in verbose or silent format (see details), default is verbose
Function for digitizing 2D landmarks on specimen images (.jpg). "nlandmarks" is the number of landmark
points to be digitized by the user. Landmarks should include "true" landmarks and semi-landmarks to be
"sliders". For best results, digitizing sequence should proceed by selecting all true landmark points first,
followed by selection of sliding semi-landmarks. Use function define.sliders.2d to select sliding semilandmarks.
If verbose=TRUE, digitizing is interactive between landmark selection using a mouse and the R console.
Once a landmark is selected, the user is asked if the system should keep or discard the selection (y/n). If "y",
the user is asked to continue to select the next landmark. If "n", the user is asked to select it again. This can
be repeated until the user is comfortable with the landmark chosen. verbose=FALSE is silent, and digitizing
of each landmark is continuous and uninterrupted until all landmarks are chosen.
Digitizing: Digitizing landmarks involves landmark selection using a mouse in the plot window, using the
LEFT mouse button (or regular button for Mac users):
Digitize the scale bar by selecting the two end points (single click for start and end),
> digitize2d("mvz206608l.jpg", nlandmarks=11, scale=3, verbose = FALSE)
Set scale = 3
Digitize each landmark with single click and the landmark is shown in red.
Select landmarks 1:11
56
When selection of n landmarks is completed, an *.nts file is created in working directory using the specimen
name, adding *.nts as a suffix.
mvz206608l.nts
"mvz206608l
1 11 2 0 dim=2
0.913794820018337 0.728605670332276
1.04745503254995 0.58684483885935
0.654575013896415 0.765058455568172
0.168537877417812 0.671901337743106
0.103732925887331 0.736706289273586
0.0551292122394711 0.813662169216032
0.103732925887331 0.927070834394373
0.330550256244013 1.04452980904337
0.686977489661655 1.00402671433682
1.03530410413799 0.927070834394373
1.71575609520803 0.643549171448521
Landmark coordinates are returned scaled.
5.2 Define sliding semilandmarks in 2D (define.sliders.2d) An interactive function to define which landmarks will "slide" along two-dimensional curves.
Function:
define.sliders.2d(spec, nsliders)
Arguments:
spec
nsliders
Name of specimen, as an object matrix containing 2D landmark coordinates
Number of landmarks to be semilandmarks that slide along curves
Function takes a matrix of digitized landmark coordinates and helps user choose which landmarks will be
treated as "curve sliders" in Generalized Procrustes analysis gpagen(). This type of semilandmark "slides"
along curves lacking known landmarks (see Bookstein 1997 for algorithm details). Each sliding
semilandmark ("sliders") will slide between two designated points, along a line tangent to the specified
curvature.
57
> data(hummingbirds)
> define.sliders.2d(hummingbirds$land[,,1], nsliders=10)
Selection: Choosing which landmarks will be sliders involves landmark selection using a mouse in the plot
window. To define the sliders, for each sliding landmark along the curve in the format 'before-slider-after',
using the LEFT mouse button (or regular button for Mac users), click on the hollow circle to choose the
landmark in the following order:
1) Click to choose the first landmark between which semi-landmark will "slide",
2) Click to choose sliding landmark,
3) Click to choose the last landmark between which semi-landmark will "slide". Selected landmarks will be
filled in and lines are drawn connecting the three landmarks, and will highlight the sliding semilandmark in
red and the flanking landmarks in blue.
example when finished,
58
This procedure is overlapping, so for example, a curve defined by a sequence of semilandmarks, the user
must select the 2nd point of the first three to be the 1st for the next e.g., 1 2 3 then 2 3 4, then 3 4 5 etc.
Function returns a .csv file to the working directory.
curveslide.csv
before slide after
1
11
12
11
12
13
13
14
15
7
15
14
12
13
14
1
16
17
16
17
18
17
18
19
18
19
20
10
20
19
2
21
22
21
22
23
22
23
24
23
24
25
8
25
24
which can be read in for use with gpagen()
> curves <- as.matrix(read.csv("curveslide.csv", header=T))
5.3 Importing 3D surface files (read.ply) A function to read ply files, which can be used for digitizing landmark coordinates or for shape warps. Other
3D file formats are currently not supported. Other files can be converted to ply using for example Meshlab
(http://meshlab.sourceforge.net).
Function:
read.ply(file, ShowSpecimen = TRUE)
Arguments:
file
ShowSpecimen
An ASCII ply file
A logical value indicating whether or not the ply file should be displayed
Function reads three-dimensional surface data in the form of a single ply file (Polygon File Format; ASCII
format only, from 3D scanners such as NextEngine and David scanners). Vertices of the surface may then be
used to digitize three-dimensional points, and semilandmarks on curves and surfaces. The function opens the
ply file and plots the mesh, with faces rendered if file contains faces, and colored if the file contains vertex
color.
> # usage, reading a file called myply.ply in the working directory
> read.ply(“myply.ply”, ShowSpecimen=TRUE)
> # view an example in the geomoprph package
> data(scallopPLY)
> myply <- scallopPLY$ply
> attributes(myply)
$names
[1] "vb"
"it"
"primitivetype" "material"
$class
59
[1] "mesh3d" "shape3d"
> plot3d(myply)
> # change color of mesh
> myply$material <- "gray" # using color word
> myply$material <- "#FCE6C9" # using RGB code
5.4 3D data collection: landmarks (digit.fixed) An interactive function to digitize three-dimensional (3D) landmarks. Input for the function is either a matrix
of vertex coordinates defining a 3D surface object or a mesh3d object as obtained from read.ply().
Function:
digit.fixed(spec, fixed, index = FALSE, ptsize = 1, center = TRUE)
Arguments:
spec
fixed
index
An object of class shape3d/mesh3d, or matrix of 3D vertex coordinates
Numeric The number landmarks (fixed, and curve sliders if desired)
Logical Whether selected landmark addresses should be returned
60
ptsize
center
Numeric Size to plot the mesh points (vertices), e.g., 0.1 for dense meshes, 3 for sparse meshes
Logical Whether the object 'spec' should be centered prior to digitizing (default center=TRUE)
Function for digitizing "n" three-dimensional landmarks. The landmarks are "fixed" (traditional landmarks).
They can be later designated as "curve sliders" (semilandmarks, that will "slide" along curves lacking known
landmarks if required. A sliding semi-landmark ("sliders") will slide between two designated points, along a
line tangent to the specified curvature, and must be defined as "sliders" using function
define.sliders.3d() or with similar format matrix made outside R. For 3D "surface sliders" (surface
semilandmarks that slide over a surface) the function digitsurface() should be used instead.
NOTE: Function centers the mesh before digitizing by default (center=TRUE). If one chooses not to center,
specimen may be difficult to manipulate in rgl window.
Digitizing: 3D Digitizing functions in geomorph are interactive between landmark selection using a mouse
(see below for instructions), and the R console. Once a point is selected, the user is asked if the system should
keep or discard the selection (y/n). If "y", the user is asked to continue to select the next landmark. If "n" the
removes the last chosen landmark, and the user is asked to select it again. This can be repeated until the user
is comfortable with the landmark chosen.
To digitize with a standard 3-button (PC):
•
the RIGHT mouse button (primary) to select points to be digitized (click-drag a box around a vertex
to select as landmark)
•
the LEFT mouse button (secondary) is used to rotate mesh,
•
the mouse SCROLLER (third/middle) is used to zoom in and out.
NOTE: Digitizing functions on MACINTOSH computers using a standard 3-button mice works as specified.
Macs using platform specific single button mice:
•
press button to rotate 3D mesh,
•
press button while pressing COMMAND key to select points to be digitized (click-drag a box around
a vertex to select as landmark),
•
press button while pressing OPTION key to adjust mesh perspective.
•
the mouse SCROLLER or trackpad two finger scroll is used to zoom in an out.
XQuartz must be configured: go to Preferences > Input > tick “Emulate three button mouse”.
NOTE: there is no pan (translate) functionality in rgl library for all platforms at this time. This is why the
function has a center=TRUE/FALSE option.
Example, reading in a mesh with 24700 vertices,
> mandible <- read.ply("Mandible.ply")
> digit.fixed(mandible, fixed=20, index=F, ptsize=1, center=T)
Select Landmark 1 #
Keep landmark 1(y/n)? #See picture step 2
y
Select Landmark 2 #See picture step 2
Keep landmark 2(y/n)?
y
Select Landmark 3 #See picture step 1
61
Function writes to the working directory an NTS file with the name of the specimen and .nts suffix
containing the landmark coordinates.
mandible.nts
"mandible
1 20 3 0 dim=3
13.2628089578878 10.2361171749175 -3.49646546534654
9.32980895788779 18.7309171749175 -1.94616546534654
-2.70691104211221 17.5351171749175 0.852234534653466
-13.7911610421122 15.6363171749175 0.945934534653464
-14.1681010421122 6.40941717491749 0.853034534653467
...
If index=FALSE (default) function returns to the console an n x 3 matrix containing the x,y,z coordinates of
the digitized landmarks. If index=TRUE, function returns a list containing a matrix containing the x,y,z
coordinates of the digitized landmarks ($selected) and a matrix of addresses for landmarks that are "fixed"
($fix, primarily for internal use).
5.5 3D data collection: landmarks and semilandmarks buildtemplate An interactive function to build template of three-dimensional surface sliding semilandmarks. Input for the
function is either a matrix of vertex coordinates defining a 3D surface object or a mesh3d object as obtained
from read.ply().
buildtemplate(spec, fixed, surface.sliders, ptsize = 1, center = TRUE)
Arguments:
spec
fixed
Name of surface file, as either an object of class shape3d/mesh3d, or matrix of three-dimensional vertex
coordinates.
The number of fixed template landmarks
62
surface.sliders The number of template surface sliders desired
ptsize
Size to plot the mesh points (vertices), e.g., 0.1 for dense meshes, 3 for sparse meshes
center
Logical Whether the object 'spec' should be centered prior to digitizing (default center=TRUE)
Function constructs a template of fixed landmarks and n "surface sliders", semilandmarks that slide over a
surface. The user digitizes the fixed points, then the function finds n surface semilandmarks following
algorithm outlined in Gunz et al. (2005) and Mitteroecker and Gunz (2009). Surface semilandmarks are
roughly equidistant set of predetermined number of points, chosen over the mesh automatically using a
nearest-neighbor approach. The set of fixed and surface slider landmarks are exported as a "template", which
is used to extract a set of similarly numbered landmarks on every specimen using function digitsurface().
Some of the "fixed" landmarks can be later designated as "curve sliders" using function
define.sliders.3d() if required - see details in digit.fixed().
To ensure a strong match between the scan and the template, it is recommended that a reasonable number of
fixed points be used. These fixed points can be designated as "curve sliders" later using function
define.sliders.3d, see the function digit.fixed() for details. NOTE: Function centers the mesh before
digitizing by default (center=TRUE). If one chooses not to center, specimen may be difficult to manipulate in
rgl window.
Digitizing: as before in digit.fixed().
Example, digitizing 16 landmarks on a mesh and then automatically calculate 150 surface sliding
semilandmarks
> myply1 <- read.ply("myply1.ply", ShowSpecimen=FALSE)
> buildtemplate(myply1, 16, 150, ptsize=1)
Select Landmark 1 #See picture step 1
Keep landmark 1(y/n)? #See picture step 2
y
63
Select all of the fixed landmarks, and then the function will calculate the position of the surface sliding
landmarks on the mesh, and plot them in blue.
Function writes to the working directory three files: an NTS file with the name of the specimen and .nts
suffix containing the landmark coordinates, "template.txt" containing the same coordinates for use with the
function digitsurface(), and "surfslide.csv", a file containing the address of the landmarks defined as
"surface sliders" for use with gpagen(). Function also returns to console an n x 3 matrix containing the x,y,z
coordinates of the digitized landmarks.
myply1.nts
"myply1
1 166 3 0 dim=3
13.2628089578878 10.2361171749175 -3.49646546534654
9.32980895788779 18.7309171749175 -1.94616546534654
-2.70691104211221 17.5351171749175 0.852234534653466
-13.7911610421122 15.6363171749175 0.945934534653464
-14.1681010421122 6.40941717491749 0.853034534653467
...
template.txt
"xpts" "ypts" "zpts"
13.2628089578878 10.2361171749175 -3.49646546534654
9.32980895788779 18.7309171749175 -1.94616546534654
-2.70691104211221 17.5351171749175 0.852234534653466
-13.7911610421122 15.6363171749175 0.945934534653464
-14.1681010421122 6.40941717491749 0.853034534653467
...
64
surfslide.csv
x
17
18
19
20
21
22
23
...
which can be read in for use with gpagen()
> sliders <- as.matrix(read.csv("surfslide.csv", header=T))
editTemplate An interactive function to remove landmarks from a 3D template file.
Function:
editTemplate(template, fixed, n)
Arguments:
template
fixed
n
Matrix of template 3D coordinates.
Number of "fixed" landmark points (non surface sliding points)
Number of points to be removed
edits a 'template.txt' file made by buildtemplate(), which
Function
must be in current working directory.
Function overwrites 'template.txt' in working directory with edited version. Use read.table("template.txt",
header = T).
> template <- as.matrix(read.table("template.txt", header=T, sep=" "))
> editTemplate(template, fixed=16, n=2)
Remove Template Points
1 of 2 points have been removed
2 of 2 points have been removed # right click and drag box around landmark
65
digitsurface An interactive function to digitize three-dimensional (3D) landmarks on a surface lacking known landmarks.
Input for the function is either a matrix of vertex coordinates defining a 3D surface object or a mesh3d object
as obtained from read.ply().
Function:
digitsurface(spec, fixed, ptsize = 1, center = TRUE)
Arguments:
spec
fixed
ptsize
center
Name of surface file, as either an object of class shape3d/mesh3d, or matrix of three-dimensional vertex coordinates.
numeric: The number of fixed template landmarks
numeric: Size to plot the mesh points (vertices), e.g., 0.1 for dense meshes, 3 for sparse meshes
Logical Whether the object 'spec' should be centered prior to digitizing (default center=TRUE)
Function for digitizing fixed 3D landmarks and placing "surface sliders", semilandmarks that slide over a
surface. Following selection of fixed points (see digitizing below), function finds surface semilandmarks
following algorithm outlined in Gunz et al. (2005) and Mitteroecker and Gunz (2009). digitsurface finds the
same number of surface semilandmarks as the “template.txt” file (created by buildtemplate()) by
downsampling scanned mesh, registering template with current specimen via GPA. A nearest neighbor
algorithm is used to match template surface landmarks to current specimen's. To use function digitsurface,
the template must be constructed first, and “template.txt” be in the working directory.
Some of the "fixed" landmarks digitized with digitsurface can be later designated as "curve sliders" using
function define.sliders.3d if required (see details in digit.fixed). NOTE: Function centers the mesh before
digitizing by default (center=TRUE). If one chooses not to center, specimen may be difficult to manipulate in
rgl window.
Digitizing: as before in digit.fixed().
Example, digitizing 16 landmarks on a mesh and fit a template to add 150 surface sliding semilandmarks
> myply2 <- read.ply("myply2.ply", ShowSpecimen=FALSE)
> digitsurface(myply2, fixed=16, pysize=1
Select Landmark 1
Keep Landmark 1(y/n)?
y
...
Select all of the fixed landmarks, and then the function will calculate the position of the surface sliding
landmarks on the mesh based on the template (blue), and plot them in green.
66
Function writes to the working directory an NTS file with the name of the specimen and .nts suffix
containing the landmark coordinates.
myply2.nts
"myply2
1 166 3 0 dim=3
13.6450089578878 9.85741717491749 -3.49156546534654
9.70490895788779 18.3504171749175 -1.90256546534653
-2.33683104211221 17.5362171749175 0.855334534653466
-14.1518710421122 15.2515171749175 1.06203453465347
-14.1681010421122 6.40941717491749 0.853034534653467
...
5.6 Define sliding semilandmarks in 3D (define.sliders.3d) An interactive function to define which digitized landmarks of an '*.nts' file will "slide" along threedimensional (3D) curves.
Function:
define.sliders.3d(spec, nsliders, surfsliders = FALSE)
Arguments:
spec
nsliders
surfsliders
Name of specimen, as an object matrix containing 3D landmark coordinates
Number of landmarks to be semilandmarks that slide along curves
If spec contains landmarks that are "surface sliders", made by buildtemplate(), "surfslide.csv" should be in
working directory
Function takes a matrix of digitized landmark coordinates, such as made by digit.fixed() or
buildtemplate() and helps user choose which landmarks will be treated as "curve sliders" in Generalized
Procrustes analysis gpagen(). This type of semilandmark "slides" along curves lacking known landmarks
(see Bookstein 1997 for algorithm details). Each sliding semilandmark ("sliders") will slide between two
designated points, along a line tangent to the specified curvature.
To define the sliders, for each sliding landmark along the curve in the format 'before-slider-after':
1) Click-drag box around landmark to choose the first landmark between which semi-landmark will "slide",
2) Click-drag box to choose sliding landmark,
3) Click-drag box to choose the last landmark between which semi-landmark will "slide", Screen will show
lines connecting the three landmarks, and will highlight the sliding semilandmark in red.
> define.sliders.3d(16,11)
67
semilandmark 9 slides between landmarks 5 and 10
This procedure is overlapping, so for example a curve defined by a sequence of semilandmarks, the user must
select the 2nd point of the first three to be the 1st for the next e.g., 1 2 3 then 2 3 4, etc.
Function returns a 'curves x 3' matrix containing the landmark address of the curve sliders, indicating the
points between which the selected point will "slide", written to the working directory as "curveslide.csv".
curveslide.csv
before
5 9
9 10
10 7
7 11
11 12
12 6
6 13
13 14
14 8
8 15
15 16
slide after
10
7
11
12
6
13
14
8
15
16
1
which can be read in for use with gpagen()
> curves <- as.matrix(read.csv("curveslide.csv", header=T))
68
Frequently Asked Questions 1) I’ve loaded my TPS file but I get a warning that no names were extracted. My TPS file has specimen
names in it, what do I do?
“ID=”
- For TPS files it is necessary to specify whether the specimen name should be read from the
line or the “IMAGE=” line. See section 1.1 for more details.
2) I want to add a groups to my data analysis, what do I do?
- Grouping variables can be made in R (e.g., c()) or imported from outside (e.g., as a .csv file).
See section 2.4 for more details.
3) How many iterations should I use in permutation tests?
- The default for geomorph functions is 999. Generally more is better, but too many can be
excessive and does not lead to more power. See section on permutation tests in the introduction for
more details.
4) How do I obtain allometry residuals for further analyses in geomorph?
- In short, you don’t. Instead, include size as a covariate. Reason: MANCOVA is correct,
MANOVA on residuals may not be (if different slopes then residuals from common slope not correct,
or if one uses residuals on each group separately then they are no longer comparable).
5) I’m using digit.fixed/buildtemplate/digitsurface. When I choose a landmark, the mesh image
disappears. And I get this error: Error in rgl.user2window(x, y, z, projection = proj) : NA/NaN/Inf in
foreign function call (arg 5). What am I doing wrong?
- For 3D digitizing in R with geomorph, one must click and drag a square around the vertex to
select it. See section 5.4 for more details and a demonstration.
69
References Adams, D. and A. Nistri. 2010. Ontogenetic convergence and evolution of foot morphology in European cave
salamanders (Family: Plethodontidae). BMC Evol. Biol. 10:216.
Adams, D. C. 1999. Methods for shape analysis of landmark data from articulated structures. Evol. Ecol. Res. 1:959970.
Adams, D. C. 2014a. A generalized K statistic for estimating phylogenetic signal from shape and other highdimensional multivariate data. Syst. Biol. In press.
Adams, D. C. 2014b. Quantifying and comparing phylogenetic evolutionary rates for shape and other high-dimensional
phenotypic data. Syst. Biol. 63:166-177.
Adams, D. C. and C. D. Anthony. 1996. Using randomization techniques to analyse behavioural data. Anim. Behav.
51:733-738.
Adams, D. C. and M. M. Cerney. 2007. Quantifying biomechanical motion using Procrustes Motion Analysis. J.
Biomech. 40:437-444.
Adams, D. C. and M. L. Collyer. 2007. The analysis of character divergence along environmental gradients and other
covariates. Evolution 61:510-515.
Adams, D. C. and M. L. Collyer. 2009. A General Framework for the Analysis of Phenotypic Trajectories in
Evolutionary Studies. Evolution 63:1143-1154.
Adams, D. C. and R. Felice. 2014. Assessing phylogenetic morphological integration and trait covariation in
morphometric data using evolutionary covariance matrices. Plos ONE 9:e94335.
Adams, D. C., F. J. Rohlf, and D. Slice. 2013. A field comes of age: geometric morphometrics in the 21st century.
Hystrix.
Adams, D. C., F. J. Rohlf, and D. E. Slice. 2004. Geometric morphometrics: ten years of progress following the
'revolution'. Italian Journal of Zoology 71:5-16.
Anderson, M. 2001. A new method for non-parametric multivariate analysis of variance. Austral. Ecol. 26:32-46.
Blomberg, S. P., T. Garland, and A. R. Ives. 2003. Testing for phylogenetic signal in comparative data: behavioral
traits are more labile. Evolution 57:717-745.
Bookstein, F. L. 1991. Morphometric tools for landmark data: Geometry and Biology. Cambridge Univ. Press,, New
York.
Bookstein, F. L. 1997. Shape and the information in medical images: A decade of the morphometric synthesis.
Computer Vision and Image Understanding 66:97-118.
Bookstein, F. L., P. Gunz, P. Mitteroecker, H. Prossinger, K. Schaefer, and H. Seidler. 2003. Cranial integration in
Homo: singular warps analysis of the midsagittal plane in ontogeny and evolution. J. Hum. Evol. 44:167-187.
Claude, J. 2008. Morphometrics with R. Springer.
Collyer, M. L. and D. C. Adams. 2007. Analysis of two-state multivariate phenotypic change in ecological studies.
Ecology 88:683-692.
Collyer, M. L. and D. C. Adams. 2013. Phenotypic trajectory analysis: Comparison of shape change patterns in
evolution and ecology. Hystrix 24:75-83.
Drake, A. G. and C. P. Klingenberg. 2008. The pace of morphological change: historical transformation of skull shape
in St Bernard dogs. Proc. R. Soc. B 275:71-76.
Dryden, I. L. and K. V. Mardia. 1993. Multivariate shape analysis. Sankhya 55:460--480.
Good, P. 2000. Permutation tests: a practical guide to resampling methods for testing hypotheses. Springer, New York.
Goodall, C. 1991. Procrustes methods in the statistical analysis of shape. Journal of the Royal Statistical Society. Series
B (Methodological):285-339.
Gower, J. C. 1975. Generalized Procrustes analysis. Psychometrika 40:33-51.
Gunz, P., P. Mitterocker, and F. L. Bookstein. 2005. Semilandmarks in three dimensions. Pp. 73-98 in D. E. Slice, ed.
Modern morphometrics in physical anthropology. Kluwer Academic/Plenum Publishers, New York.
Gunz, P., P. Mitteroecker, S. Neubauer, G. W. Weber, and F. L. Bookstein. 2009. Principles for the virtual
reconstruction of hominin crania. J. Hum. Evol. 57:48-62.
Kendall, D. G. 1984. Shape-manifolds, Procrustean metrics and complex projective spaces. Bulletin of the London
Mathematical Society 16:81-121.
Klingenberg, C. P. 2009. Morphometric integration and modularity in configurations of landmarks: tools for evaluating
a priori hypotheses. Evol. Dev. 11:405-421.
Klingenberg, C. P., M. Barluenga, and A. Meyer. 2002. Shape analysis of symmetric structures: quantifying variation
among individuals and asymmetry. Evolution 56:1909-1920.
70
Klingenberg, C. P. and N. A. Gidaszewski. 2010. Testing and quantifying phylogenetic signals and homoplasy in
morphometric data. Syst. Biol. 59:245-261.
Klingenberg, C. P. and G. S. McIntyre. 1998. Geometric morphometrics of developmental instability: Analyzing
patterns of fluctuating asymmetry with procrustes methods. Evolution 52:1363-1375.
Mardia, K. V., F. L. Bookstein, and I. J. Moreton. 2000. Statistical assessment of bilateral symmetry of shapes.
Biometrika 87:285-300.
Mitteroecker, P. and P. Gunz. 2009. Advances in Geometric Morphometrics. Evol. Biol. 36:235-247.
Mitteroecker, P., P. Gunz, M. Bernhard, K. Schaefer, and F. L. Bookstein. 2004. Comparison of cranial ontogenetic
trajectories among great apes and humans. J. Hum. Evol. 46:679-697.
O'Higgins, P. and N. Jones. 1998. Facial growth in Cercocebus torquatus: an application of three-dimensional
geometric morphometric techniques to the study of morphological variation. J. Anat. 193:251-272.
Rohlf, F. J. 1998. On applications of geometric morphometrics to studies of ontogeny and phylogeny. Syst. Biol.
47:147-158.
Rohlf, F. J. 1999. Shape statistics: Procrustes superimpositions and tangent spaces. J. Classif. 16:197-223.
Rohlf, F. J. 2002. Geometric morphometrics and phylogeny. Pp. 175-193 in N. MacLeod, and P. L. Forey, eds.
Morphology, shape and phylogeny. Francis & Taylor, London.
Rohlf, F. J. 2010. tpsRelw: Relative warp analysis. Version 1.49, Department of Ecology and Evolution, State
University of New York at Stony Brook, Stony Brook, NY.
Rohlf, F. J. 2012. NTSYSpc: Numerical taxonomy and multivariate analysis system. Version 2.2. Exeter Software,
New York.
Rohlf, F. J. and L. F. Marcus. 1993. A revolution in morphometrics. Trends Ecol. Evol. 8:129-132.
Rohlf, F. J. and D. Slice. 1990. Extensions of the Procrustes Method for the Optimal Superimposition of Landmarks.
Syst. Zool. 39:40-59.
Zelditch, M. L., D. L. Swiderski, and H. D. Sheets. 2012. Geometric morphometrics for biologists: a primer. Elsevier,
Amsterdam.
71